The actual sensitivity of each indicated ISO is measured using the same shots as are used to measure ISO noise levels, we simply compare the exposure for each shot to the metered light level (using a calibrated Sekonic L-358), middle gray matched. We estimate the accuracy of these results to be +/- 1/6 EV (the margin of error given in the ISO specifications). Note that these tests are based on the sRGB JPEG output of the cameras, in accordance with ISO 12232:2006, the standard used by camera manufacturers.
By our tests, the X100S's measured sensitivities are within 1/6EV of indicated, which is within the tolerance allowed by the ISO specification. In other words, ISO 100 indicated = ISO 100. Interestingly, when we looked at the X-Pro 1 and X-E1, we found that their measured ISO sensitivities are about 1/3 to 1/2 stop lower than marked. This discrepancy (which has little importance for real-world photography) explains why the X100S's measured noise in Raw mode appears higher than these cameras overall (see graphs below). In JPEG mode, the X100S's measured noise is a little higher than the X-Pro 1 and X-E1 at lower ISO sensitivities, and a fraction lower at the very highest ISO sensitivities. This is due to changes that Fujifilm has made to its noise-reduction algorithms - changes which again, make a negligible difference in everyday photography.
Noise and Noise Reduction (JPEG)
Up to ISO 3200, images from the X100S display very little noise, to the extent that in normal shooting, we don't really worry about noise very much at all. At ISO 6400 noise is noticeable, but doesn't have much of an impact on critical image quality, and it is only at ISO 12,800 and 25,600 that detail is seriously compromised. Turning noise reduction down has the expected effect of increasing visible 'graininess' and slightly improving detail resolution, but the effect is not dramatic. Likewise turning noise-reduction up. You lose a tiny bit of definition, and midtones are a little smoother at ISO 3200 and above, but the effect is subtle.
If we turn our attention to the measured noise graph, we can see that the X100S gives higher noise measurements than the X-E1 (and X-Pro 1) up to ISO 6400, and slightly lower at ISO 25,600 (which along with ISO 12,800 is a JPEG-only sensitivity). This suggests there are two factors at play - the revised ISO sensitivity (detailed at the top of this page), which will slightly disadvantage the X100S in this test and what appears to be a tweak to the X100S's noise-reduction algorithms compared to its X-Trans cousins. The perceptual differences are pretty small though, with the X100S doing a similar job to the other X-series cameras in terms of retaining detail and keeping noise at bay.
Adobe Camera Raw noise (ACR 8.1)
Here we look at the Raw files processed through Adobe Camera Raw (in this case version 8.1). Images are brightness matched and processed with all noise reduction options set to zero. Adobe does a degree of noise reduction even when the user-controlled NR is turned off.
The amount of NR applied 'under the hood' is not high, but it does vary by camera (Adobe is attempting to normalize output across different sensors), so inevitably we are still looking at a balance of noise and noise reduction, rather than pure noise levels. However, the use of the most popular third-party raw converter is intended to give a photographically relevant result, rather than simply comparing sensor performance in an abstract manner.
As with the the X-E1, Adobe Camera Raw's processing of the X100S's files looks radically different compared to its output from conventional Bayer sensor cameras. Chroma noise is strikingly low, and detail retention is impressively high - very much like the camera's JPEGs, in fact. Because of this, direct comparisons have to be treated with a degree of caution - it's best to assume that the de-mosaicing process of the X-Trans CMOS sensor behaves as though it's doing substantial chroma noise reduction relative to ACR's standard treatment of Bayer sensors.