I undertook a personal search in an attempt to try and find the limits and conditions where one could expect the EOS 5D Mark 3’s light leak to affect their metered exposure. In the process of making these measurements, I’ve tried to account for and control errors as best I can; however, in many cases I lack the appropriate instrumentation to make precise measurements.

My initial objective had been to build a mathematical model to quantify the problem and then make testable predictions from it. In the process of doing that it became painfully clear that I lacked the precision in test instrumentation to make it count for much. Though it did turn out to be a good point to start from in the empirical tests I ended up doing.

A Word About Outliers, Bias, and Error

I don’t work for Canon, in some ways I wish I did, as I’d at least have a paycheck coming in for this kind of thing. That also means that I have a completely statistically irrelevant sample size of 1 camera. The one I bought.

That brings me to outliers. I don’t think my camera is one, either good or bad. However, without a statistically relevant sample size, I can’t say for sure. What I can say for sure, is if your 5D mk. 3 shows consistent exposure errors under normal shooting conditions in daylight or otherwise, there’s something wrong with your camera.

One other thing, in this article unless I say otherwise assumes when you see EV I mean EV₁₀₀, or the EV at ISO 100.

The Good News

The good news is that, in my opinion, it takes a pretty impressive set of hoops to jump through to get the meter to start being a problem, and especially a more than 1/3^rd stop problem. Moreover, in very low light situations you can go quite a bit lower than 1 EV and still not see the backlight affect the exposure, especially with an f/1.4 lens.

Alone in the Dark with Only a Backlight to Guide You

In my testing, with a Canon EF 24-70mm f/2.8L USM lens, it took a scene somewhere around  -3 and -2-2/3rds EV to induce a backlight error. With a 24-105mm f/4L IS USM, that jumped a bit to around -1-2/3 to -1-1/3 EV.

The difference between the two lenses is a byproduct of their various transmission percentages (the 24-70/2.8L is a t/3.3 lens wide open, the 24-105/4L is a t/5.1 lens at 70mm wide open). The difference in transmission is a bit less than 1-1/3^rd stops and that nicely corresponds to the differences in scene illuminances where the error first occurred, as it should.

This brings up an important point, and it applies to more than just the 5D mark 3, but in fact to all camera meters. The speed of the lens, specifically in t-stops, will affect the bottom limit for the meter’s accuracy. A 24mm f/1.4L will be able to meter accurately in a darker scene than a 17-40 f/4L will. This is simply a byproduct of the fact, that the lens always attenuates some light before it can reach the sensor or meter. The light the lens attenuates, the more noise, or in the case of the 5D-3 leaked light, will influence the meter readings.

Incident Light: Think Flashlight not Backlight

The problem isn’t strictly related to the backlight, that’s just what prompted its discovery. More generally, the problem is that light on the top LCD or produced by the backlight for the top LCD leaks to the meter. The amount that does is a very small percentage of the actual incident light; however, given the right conditions it can become significant enough to affect exposures in other than just dark situations.

This is the not so good news for some, and a non-issue for others, and much, much, more complicated than just the backlight in the dark.

Because of the nature of the leak, what we’re really looking at is a situation where the meter is fed light from a 3^rd highly attenuated source, in addition to the lens and viewfinder. How attenuated I’m not entirely sure yet, but it looks likely to be about 35,000-100,000 times. This works out to something like 15.1-16.5 stops (including the losses of the lens), give-or-take, to show a 1/3^rd stop error.

This isn’t necessarily great but it does mean that uncontrolled illumination of the LCD isn’t going to be an appreciable problem for most people. With the camera in direct sunlight, it takes about 10 stops of ND + a t/5+ lens to show a 1/3 stop error in metering. I’ve done 3 direct tests of this, metering a clear blue northern sky as a similarly lit approximately mid toned target.

There is a caveat to all of this though, included in that 15-odd stops, is the lens’s transmission as well. In other words, a person using an f/1.4 lens will be able to operate over a wider range of conditions than someone using an f/5.6 lens will. Likewise, using an adapted lens where stop-down metering is necessary will be even further constrained.

In practical terms, you have to place a considerable amount of neutral density in front of the camera’s lens, or have an incredibly slow lens, to prompt a metering error in daylight.

Conclusions

The one real conclusion that I can draw from this is that, as Canon states, the light leak won’t be an issue for most users. Interestingly, the people most likely to be affected are actually ones that are using lots of ND and still expecting the camera to meter though the ND instead of metering and adjusting for the ND manually; especially if they are using adapted lenses that have to be manually stop-down metered.

Moreover, in the time since I started looking at the problem and now, Canon has announced that they’ll fix cameras for people that believe this is a problem for them, as well as having updated their manufacturing to include a better light shield over the meter and prism.

What I can say from my own experience, and I’ve done a considerable bit more shooting with the camera in practical situations since I looked at the problem, is that there’s very little if any reason for most people to run back to Canon to get this fixed, or even bother. While I may have this addressed eventually on my body, I’m in no rush to do so either and will definitely only be sending my camera in when it’s convenient for me.