Camera Sensor Size
& Pixel Size

1. Sensor Formats, Pixel Size and DXOMark Scores — Sorted by Pixel Pitch

All entries sorted from largest pixel pitch to smallest. Sensor dimensions, megapixel count, pixel pitch, and verified DXOMark scores are shown in a single table. DXOMark scores are measured at base ISO. Rows highlighted in yellow have a pixel pitch of 5 µm or larger. Cameras with no DXOMark score are noted with —.

★ ≥26 bits = colour depth of 26 bits or above at base ISO (top tier).   CCD = Charge-Coupled Device sensor. CMOS = Complementary Metal-Oxide Semiconductor sensor (includes BSI variants). Blue ADC values = 16-bit sensors. All others are 14-bit. 16-bit raises the theoretical colour space ceiling from 42 bits (14+14+14) to 48 bits (16+16+16) — though noise still determines the usable portion.

† Fujifilm X-Trans sensors use a non-Bayer colour filter array that DXOMark cannot process — no scores available.   ‡ The Sony A7S III's IMX510 sensor is a 48MP Quad-Bayer design with real photosites at ~4.2 µm pitch. Sony combines four same-colour photosites digitally, after readout, into each 12MP output pixel — the "~8.4 µm pixel" is an effective/binned figure, not a genuine single large photosite (confirmed by sensor teardown/microscopy; see §3 and §4 for what this changes).

Not yet tested: Panasonic S1R II, Leica SL3-P, Canon EOS R5 Mark II, Canon EOS R6 Mark II, Nikon Z6 III, Nikon Zf, Nikon Z5 II, OM System OM-1 / OM-1 Mark II.

The largest full-frame pixel pitch in this table — and the largest ever shipped in a 35mm-format production camera — belongs to the Contax N Digital. Released in 2002 with a Philips FTF3020-C CCD, its ~6MP sensor produced 12.0 µm pixels, roughly a third larger than the Phase One P25+ and nearly half again the size of the Sony A7S III's. It predates DXOMark testing (founded 2008), so no verified colour depth, dynamic range, or low-light ISO score exists, but it holds a unique place in the table as the only full-frame 35mm CCD sensor ever to reach production — both Canon and Kodak's later full-frame cameras of the era used CMOS instead. Its analog-to-digital conversion was 12-bit — the official Contax N Digital specification lists RAW capture as "RAW (12 bit)," consistent with the 2002-era CCD readout chain (12-bit converter), and the lowest ADC depth in this table. The camera was commercially short-lived, withdrawn within a year, but contemporary and retrospective reviews consistently single out its colour rendition and tonal richness, consistent with the large-pixel/CCD advantages discussed in Section 2.

The Phase One P25+ uses a CCD sensor with ~9.0 µm pixels and a 16-bit ADC, mounting on Hasselblad 500-series V bodies. The Sony A7S III follows at an effective 8.4 µm with a 48MP Quad-Bayer BSI-CMOS sensor binned 2×2 to a 12MP output — its real photosites are ~4.2 µm, larger than any current-production medium format camera in this table only in the binned/effective sense. The Leica SL-system spans four pixel size tiers: 6.9 µm (M8/M9 CCD), 6.0 µm (M240/M10), 4.43 µm (SL3-P), and 3.76 µm (M11/Q3/SL3). The new SL3-P shares its 44MP BSI sensor platform with the Panasonic Lumix S1R II. The Canon R3 and Leica M240/M10 share the same ~6.0 µm pixel pitch. The Pentax K-1 Mark II's ~4.88 µm full-frame pixels show that large pixel design is not exclusive to mirrorless cameras.

2. Why Larger Pixels Produce Richer Colour

Photographers consistently notice that cameras with larger pixels produce more vivid, saturated, and deeper colours. There are four clear physical reasons.

2.1 More Photons Per Pixel — Better Signal-to-Noise Ratio

A larger photosite collects more light. With more photons, the signal is much stronger relative to the noise floor. Colour information lives in the differences between R, G, and B channels — noise corrupts those differences. Clean channels mean vivid, accurate colour.

2.2 Less Colour Bleeding Between Channels

Smaller pixels are physically closer together, so light scatters sideways between neighbouring photosites (optical cross-talk), mixing colour channels and reducing colour purity.

2.3 Dynamic Range and Tonal Depth

Larger pixels hold more electrons before clipping (larger well capacity), giving smoother tonal gradations and richer colour across both shadows and highlights simultaneously.

2.4 Less Chroma Noise Suppression Needed

Raw converters aggressively reduce chroma noise on small-pixel sensors, literally desaturating fine colour detail. Large pixel sensors need far less of it.

3. Why Larger Pixels Do Not Always Win — The A7S III Case

The Sony A7S III is marketed and widely discussed as an 8.4 µm-pixel sensor — the largest BSI-CMOS in the full-frame table — yet its base ISO colour depth is lower than the Nikon D850's, which has real 4.35 µm pixels. The comparison looks even more puzzling once you know the A7S III isn't actually a large-photosite design at all: its IMX510 sensor is a 48MP Quad-Bayer array with genuine photosites at ~4.2 µm, digitally combined 2×2 after readout into a 12MP output (see ‡ note, Section 1). Understanding that architecture — not a simple "big pixel vs. small pixel" framing — is what actually explains its colour depth score.

3.1 What Drives Colour Depth

DXOMark's actual colour depth metric is not a simple formula — it's a noise-covariance integral over the full 3D RGB colour space (see §5.2). But the same four physical factors that set a sensor's dynamic range also set the ceiling on its colour depth, since both derive from how much signal clears the noise floor:

Dynamic range (a proxy for the colour-depth ceiling):
  DR (stops)  ≈  log2(Full_Well / Read_Noise)

Four factors determine the result:
1. Full well capacity  — electrons before clipping
2. Read noise          — electronic noise of readout circuit
3. Quantum efficiency  — photons that become electrons
4. Architecture        — BSI/FSI, CCD/CMOS, readout speed

This ratio is the standard single-channel dynamic-range formula, not DXOMark's colour-depth formula — but a sensor with a higher dynamic-range ceiling from these same factors will generally also score a higher colour depth, which is why the two numbers move together in practice.

These four factors describe what physically limits dynamic range and colour depth — but they are not four independent knobs a camera designer trades against each other per model. Full well capacity, quantum efficiency, and the base CCD/CMOS/BSI architecture are fixed by the sensor die itself: they're identical across every camera built on that die and can't be selectively "sacrificed" without literally using a different sensor. The one choice that genuinely is made and tuned per device is the colour filter array (CFA) — the dye/filter transmission profile, especially how much light the green filter passes through. A more transmissive CFA raises overall sensitivity (a lower/faster native or base ISO) at the cost of channel separation, which pushes more correction onto the white-balance and colour-matrix gains applied afterward — and those gains amplify noise (see §5.2). So the real per-device trade-off is native ISO vs. native white-balance/colour accuracy, not the four physical factors above. Read noise has a smaller secondary tunable component: supporting electronics — the analog front end, ADC design, and on-chip processing — can measurably improve it even on an unchanged sensor die. The Sony A7S and A7S II demonstrate this directly: Sony's own engineers stated both cameras share the identical 12.2MP Exmor CMOS sensor and processing engine, yet the A7S II's revised circuitry and noise-reduction pipeline changed real-world noise behaviour while DXOMark measured dynamic range as essentially unchanged and colour sensitivity only ~0.3 bits apart — full well capacity, quantum efficiency, and architecture stayed exactly the same; only the surrounding electronics did.

3.2 What Quad-Bayer Binning Actually Trades Away

The D850 gets its colour depth from one real photosite per output pixel. The A7S III gets its 12MP output by digitally summing four real ~4.2 µm photosites per output pixel — after each has already been read out and converted. That distinction matters because combining digital reads sums their noise variance: four independent read noise sources add up, so the combined read noise rises by roughly √4 (2×) rather than being averaged down the way true analog binning or a genuinely larger single photosite would allow. The light-gathering side still wins — four real photosites collect roughly four times the photons of one, so photon shot noise (which scales with signal) is still much better controlled than on a native 12MP sensor with 4.2 µm pixels would be — but the read-noise penalty from digital-domain binning is the real reason the A7S III doesn't reach the D850's colour depth despite its large effective pixel pitch.

4. Why the A7S III Image Can Look Better Than the D850 in Practice

Despite a lower DXOMark colour depth score, many photographers find the A7S III produces more pleasing images. Five reasons explain why.

4.1 You Are Rarely Shooting at Base ISO

DXOMark colour depth scores are measured exclusively at base ISO. Above ISO 400–800 the A7S III's binned photosites — four real ~4.2 µm photosites summing their light per output pixel — start pulling ahead dramatically, since the summed photon capacity keeps shot noise low even as the D850's smaller native photosites run out of signal. At ISO 6400 it is in a completely different league. Most real shooting happens at elevated ISOs where the D850 struggles and the A7S III still produces clean, rich colour.

4.2 Noise Reduction Quality vs. Quantity

The A7S III produces so little noise that it needs almost none applied. The D850 at the same ISO requires visible noise reduction which softens micro-contrast and desaturates colour transitions. A lower DXOMark score with zero noise reduction applied often looks richer in practice.

4.3 Resolution Advantage Disappears at Normal Output Sizes

Colour rendering does not degrade when a high-resolution image is displayed smaller — downsampling a 45MP file to screen size averages many pixels together, which actually suppresses noise and improves apparent colour quality. The A7S III's colour advantage is baked in at the moment of capture — its 2×2 binning already happens on-sensor, at the photosite level, not at the display stage. What disappears at normal output sizes is the D850's resolution advantage — you cannot see more detail in a screen-sized image than the A7S III already provides. The colour and tonal quality of the A7S III hold up fully at any output size because they originate in sensor physics: real photosite size and count, summed full well capacity, binning-related read noise, and architecture.

4.4 Colour Rendering Is Not Just Colour Depth

DXOMark measures how many colours are distinguishable — not how pleasing they are. Sony's colour science tuning may produce more pleasing skin tones or richer blues independently of the bit depth score.

4.5 The 6.5× Figure Assumes Identical Conditions

The mathematical comparison only holds at base ISO, same exposure, same raw converter, zero noise reduction. Change any one — particularly ISO — and the result can reverse entirely.

5. The Two Meanings of Bit Depth

5.1 ADC Bit Depth — Hardware Output

How many bits the analog-to-digital converter outputs per pixel. A hardware design choice, not a quality measurement. The Phase One P25+ uses a 16-bit ADC — two bits more than most modern cameras.

10-bit ->  1,024 levels per channel  (typical smartphone RAW)
12-bit ->  4,096 levels per channel
14-bit -> 16,384 levels per channel  (most modern cameras)
16-bit -> 65,536 levels per channel  (Phase One P25+, IQ4, some MF backs)

5.2 DXOMark Colour Depth — Measured Quality Score

Not the ADC bit depth. Per DXOMark: "since each channel has its own noise level, it actually distinguishes far fewer than the theoretical bits of RAW data." The score measures truly distinguishable colours after noise is accounted for, across all three channels combined.

5.3 The 14+14+14 Explanation

Theoretical maximum colour space:
  R channel: 14-bit  =  16,384 levels
  G channel: 14-bit  =  16,384 levels
  B channel: 14-bit  =  16,384 levels
  Total:     14 + 14 + 14  =  42 bits of 3D colour space

What noise does to that:
  DXOMark measured score  =  ~23–26 bits  (usable portion)
  Destroyed by noise      =  ~16–19 bits  (unusable)
  Noise eats roughly half the theoretical colour information.

For a 16-bit sensor (Phase One P25+, IQ4):
  Theoretical maximum  =  16 + 16 + 16  =  48 bits
  The 16-bit ADC raises the ceiling, but noise still
  determines how much of that ceiling is actually usable.

Why does a 14-bit camera score above 14 bits? Because the score covers the combined three-dimensional R×G×B colour space, not per channel. The theoretical maximum for a perfect 14-bit sensor would be 42 bits. Real sensors achieve ~23–26 bits because noise reduces each channel's effective precision.

5.4 Leica in Context — A System That Went the Other Way

The Leica SL-system now spans three distinct pixel size tiers. The M8 and M9 used CCD sensors with ~6.9 µm pixels — enormous for their era, highly valued for colour richness and tonal smoothness. The M240 and M10 moved to CMOS with ~6.0 µm pixels, still very large. The M11, Q3, and SL3 generation dropped to 3.76 µm driven by the jump to 60MP — yet the M11 achieves 26.3 bits colour depth and 14.8 EV dynamic range, matching the Nikon D850 exactly, thanks to BSI architecture and a base ISO of 64. The newest addition, the SL3-P (announced June 2026), takes a different path: its 44MP BSI sensor gives ~4.43 µm pixels — sitting alongside the Sony A7R V and Canon R5, and sharing its sensor platform with the Panasonic Lumix S1R II. DXOMark scores are not yet available for the SL3-P.

5.5 The Phase One P25+ in Context

The P25+ represents an extreme endpoint of the pixel size spectrum among medium-format cameras — ~9 µm pixels on a CCD sensor with a 16-bit ADC. CCD sensors have very low fixed pattern noise and excellent colour accuracy, but higher read noise at the circuit level. At base ISO in good studio light the P25+ produces files with exceptional tonal richness. Its maximum native ISO is around 400 and it is a pure studio tool. Among 35mm-format cameras specifically, the older Contax N Digital (Section 1) exceeds it in pixel pitch at 12.0 µm, though it has no comparable measured data.

5.6 Why the A7S III Has a Lower Low-Light ISO Score Than the Z6

The DXOMark low-light ISO score is the highest ISO where all three thresholds are simultaneously met: SNR >= 30 dB, Dynamic Range >= 9 EV, and Colour Depth >= 18 bits. The A7S III's colour depth drops below 18 bits sooner than the Z6's as ISO rises. Beyond ISO 3200 the A7S III produces measurably cleaner images than any other full-frame camera — the headline score does not capture this because it is a threshold test, not a continuous measurement.

5.7 Why Dynamic Range Can Exceed ADC Bit Depth

DXOMark normalises to 8MP equivalent output. Averaging 45MP down to 8MP reduces noise by cancelling random variation across multiple pixels, pushing the noise floor down and widening the measured dynamic range beyond what any single pixel achieves. This is why 14.8 EV is measurable from a 14-bit ADC.

5.8 Colour Depth Is Not SMI — Two Different "Colour Quality" Numbers

It's easy to conflate DXOMark colour depth with a completely different metric that also gets called "colour" quality: the Sensitivity Metamerism Index (SMI), defined in ISO 17321. They answer different questions:

Colour depth (DXOMark)  ->  HOW MANY colours can the sensor tell apart?
                            A noise-limited count, in bit-equivalent units (see §5.2).
                            Says nothing about whether those colours are *correct*.

SMI (ISO 17321)         ->  HOW ACCURATELY does the sensor see the colours it captures,
                            compared to the human eye?
                            A 0-100 score (occasionally negative). SMI = 100 means the
                            sensor's spectral response is a perfect linear combination of
                            the eye's cone responses - it never confuses two colours
                            (metamers) that look identical to a human but have different
                            spectra, or vice versa. Typical DSLR/mirrorless SMI is ~75-85;
                            phone cameras are often around 40.

A sensor can have excellent colour depth (lots of noise headroom, many distinguishable colours) and mediocre SMI (systematically mis-seeing some real-world colours relative to human vision), or the reverse. Neither DXOMark's public camera-sensor rankings nor this document report SMI figures — if a specific colour rendition "just looks right" to you on one camera in a way that doesn't track its DXOMark colour-depth score, metamerism matching (SMI) rather than noise-limited colour depth is a plausible reason why, alongside the colour-science/tuning point already made in §4.4.

6. Highlight Headroom and Rolloff — Comparing Sensors Near Clipping

This is one of the most practically important sensor characteristics and is not captured by any single DXOMark headline score.

6.1 What Happens Near Clipping

Sensors do not clip suddenly like a hard wall. The response curve gradually compresses before full saturation — called the shoulder of the photographic response curve, borrowed from film terminology. Its shape determines whether blown highlights look harsh and digital or smooth and film-like.

Hard clipping:  Signal ─────────────────/   (clips abruptly)
Soft rolloff:   Signal ─────────────────╮   (gradual compression)
                                         ╰─  (soft clip)

6.2 What the Measured Data Tells Us

The D850's 14.8 EV vs A7S III's 13.9 EV means the D850 has approximately 0.9 EV more total tonal range. Testing shows the D850 uses most of this in shadow recovery. The A7S III's large pixels give a gentler highlight shoulder despite the narrower EV range — its rolloff is often described as more filmic. The CCD sensor in the Phase One P25+ has a naturally soft highlight rolloff, one reason older CCD medium format backs remain valued for portrait and product work.

6.3 How to Measure Highlight Headroom

photonstophotos.net: Bill Claff's interactive PDR chart shows full ISO curves for all cameras. Select any two and see exactly where each begins compressing and where it hard-clips at every ISO setting.

Practical RAW test: Shoot a bright scene overexposed by 1, 2, and 3 stops. Open raw files with zero processing. Pull exposure down by the same amount. Compare what detail survives and how the transition to clip looks.

RawDigger channel analysis: Plot mean pixel value per channel against exposure. Where the curve departs from linearity is the start of the shoulder. Distance to hard clip is usable headroom.

7. Signal-to-Noise Ratio (SNR)

7.1 Definition

SNR = Signal / Noise
Signal = mean pixel value at a given exposure
Noise  = standard deviation of pixel values at the same exposure

SNR is a curve, not a single number — it changes at every exposure level. DXOMark's low-light ISO threshold is the highest ISO where SNR >= 30 dB, dynamic range >= 9 EV, and colour depth >= 18 bits simultaneously.

7.2 Practical RAW Measurement

1. Hard clip value  =  62,000  (just below 16-bit max of 65,535)
2. Black floor      =  100     (darkest region, no light)
3. Linear ratio     =  62,000 / 100 = 620 levels
4. To stops         =  log2(620) = log(620)/log(2) = 9.28 stops
5. To dB            =  9.28 × 6.02 = 55.8 dB

1 stop = 6.02 dB  because  20 × log10(2) = 6.0206

Subtract the manufacturer pedestal from the black floor value for greater accuracy.

7.3 Five Measurement Methods

8. Dynamic Range vs. SNR

Dynamic Range is always larger than maximum SNR.

Dynamic Range  =  Full_Well / Read_Noise
SNR_max        =  sqrt(Full_Well)

DR > SNR_max when: sqrt(Full_Well) > Read_Noise — always true in real sensors

Example (Full_Well = 50,000 e-, Read_Noise = 5 e-):
  DR      = 50,000 / 5    = 10,000  →  80 dB
  SNR_max = sqrt(50,000)  = 224     →  47 dB

Dynamic Range benefits from both low read noise and high full well capacity. Maximum SNR depends only on electrons collected — they reward different physical properties of the sensor.

9. Where to Find Verified Sensor Measurements

DXOMark (dxomark.com)

Portrait score = colour depth in bits. Landscape score = dynamic range in EV. Sport score = low-light ISO. Valid for cameras with RAW output. Smartphone scores use a different protocol and are not directly comparable. All headline scores are at base ISO only.

Bill Claff — photonstophotos.net

The most trusted independent source. Full dynamic range curves across all ISOs, read noise in electrons, full well capacity — all with fully transparent methodology. Essential for understanding highlight headroom and performance across the entire ISO range.

Self-measurement with RawDigger

The highest-trust measurement — your own camera at your actual shooting conditions. Reads RAW files before any processing.

10. Summary

The largest pixel pitch in this entire table belongs to the Contax N Digital at ~12.0 µm — a 6MP full-frame CCD from 2002 and, to date, the largest pixel ever fitted to a 35mm-format production sensor. It has no DXOMark-era data, so it sits outside the measured comparisons, but it sets the practical ceiling that everything else in the table is measured against.

Large pixels provide better colour depth all else being equal — but all else is rarely equal, and raw pixel pitch alone isn't even a fair comparison once you account for what's actually verified. The Phase One P25+ sits second only to the Contax N Digital in raw pixel pitch at ~9 µm — a CCD sensor with 16-bit capture designed for studio use — but, like the Contax N Digital, it has no verified DXOMark score (the — entries in the Section 1 table). Phase One's related, larger P65+ back was tested at 26.0 bits / 13.0 EV, but that's a different sensor from the P25+ and shouldn't be read as a stand-in for it. Among cameras that do have verified scores, the Sony A7S III/A7S II pair sits at the top of the pixel pitch table at an effective 8.4 µm — the A7S III really a 48MP Quad-Bayer BSI-CMOS sensor with ~4.2 µm native photosites binned 2×2, the A7S II a genuine native 8.4 µm design — yet both score a lower base ISO colour depth than the Nikon D850's genuinely large 4.35 µm photosites. These cameras illustrate why pixel size alone never tells the complete story, and why "pixel pitch" needs to specify native vs. binned/effective, and verified vs. unverified, before it means anything.

The verified DXOMark data reveals two surprises: the Leica M11's 60MP BSI-CMOS sensor achieves 26.3 bits and 14.8 EV — matching the Nikon D850 on dynamic range despite much smaller pixels, thanks to BSI architecture and a base ISO of 64. And the A7S III's low-light ISO score (2520) is lower than the Nikon Z6 (3299) despite being a far superior low-light camera.

Laboratory measurements and practical experience are both valid. They measure different things. Understanding what each one measures — and what it does not — is what allows you to choose the right camera for your actual work.

11. Resources and Further Reading