A scatter plot of ChIP-chip data versus mock control data. Data from the X (red) and the 2L (blue) chromosome were superimposed after median smoothing (window size 7) to suppress background noise. Choosing a threshold solely based on the experiment is equivalent to drawing a horizontal line, and it is clearly not optimal for separating the signal from the background. A direct subtraction of mock control from the experiment corresponds to using a line with the slope 1 as a threshold (the blue line or another parallel to it), but that also does not appear to be optimal. A better solution would be the red line, which corresponds to a subtraction with a different weight for the control.