In the heat of FF vs. DX debates the fact that it is about digital sometimes is forgotten. Statements like "smaller sensor needs more magnification" sound very confusing. But there is yet another aspect of the problem - demosaicing. Here is a typical example of how DoF on film is explained: "Imagine the very tip of a pin with a size of exactly zero, located precisely in the plane of perfect sharpness, that means, it is imaged to the film with a size of exactly zero, not widened by any blur. Now, move this pin toward the camera and watch the diameter of its tip increase by blurring. When it has reached 30 μm, halt the pin! It is now right at the inner border of the depth of field. Now, do the same in the opposite direction. Beyond the plane of perfect sharpness you will reach the outer border of the depth of field." (from Zeiss newsletter CLN #1, p.3, article "Depth of Field – An Insider’s Look Behind The Scenes") Film has very fine grain, but often grains are overlapping, forming nearly continuous sensitive surface. Not so with digital sensors. Moreover, there are substantial gaps between cells, and 5mkm cell can have only 3mkm sensitive surface. Demosaicing interpolate from the cell to all neighboring cells, making luminance and colour data to distribute to at least 8 cells (3x3 square). That means our imaginary 5mkm sensor cell allows to record data in 15 μm linear. (Digression: to complicate things even more, influence of data from a cell is attenuated in demosaicing depending on the distance, and as a result we have blur. The larger are the cells, the more blur we have. This is one of the reasons why smaller pixel count on a smaller sensor can provide images sharper then bigger sensor with greater pixel count, even with AA filters being taken out.) Now, if we try to apply the explanation given above to the digital sensor... Let's assume the tip of a needle is lucky to be projected right into the center of a pixel. When we start moving it, the tip immediately jumps to effective size of the pixel, and thus is demosaiced in 8 surrounding pixels, forming 3x3 square - linear size of 15 μm. Next jump will be to the size of 3 pixels, and demosaicing will extend it to 25 μm. If the tip is located not at the center of a pixel cell, but between cells, its first race will be 10 μm, and it will be blurred by demosaicing into 20 μm. If no pixel-level sharpening is appied to the image in post-processing, and we want our CoC in the resulting image to be 20 μm (30 μm/1.5, that is 4 pixels of the sensor), for taking the image we need CoC of 10 μm. If we apply sophisticated demosaicing, which restores luminance information in each pixel using multi-pass approach, or even simple pixel-level USM (400..500/0.2..0.3/0..1), we are getting pretty much to cell size of CoC.