Thank you, Paul and Hidong, for your explanations. Here is another way of looking at it (kindly provided by Ronald Stenkamp).
=== Think of the electron density as a 3-dimensional function with an average value of 0.0 (This is true if you have not included an F000 reflection, and it's true of difference electron density maps). You can take that function and calculate its rms value. That would be its rms deviation from the average, and you can convert that to an estimated standard deviation (or simply call it because of the large number of data points in this function). Sigma is the standard deviation, and it's a quantitative way of assessing the noise level of the map. So you can then ask the following question for any peak in the map: Is this peak significant or not? One way to decide on that is to ask how much larger is this peak than the estimated standard deviation of the map? High peaks, because they are much above the noise, are more significant than are the low peaks. And high peaks are those that will be shown on your graphics screen as you increase the sigma level of the contours. ===
