|Recent studies have reported a reversal of an earlier trend in income segregation in metropolitan regions, from a decline in the 1990s to an increase in the 2000-2010decade. This finding reinforces concerns about the growing overall income inequality in the U.S. since the 1970s. We re-evaluate the trend. Because the effective sample for the ACS is much smaller than it was for Census 2000, to which it is being compared, there is a possibility that the apparent changes in disparities across census tracts result partly from a higher level of sampling variation and bias due to the smaller sample. This study uses 100% microdata from the 1940 census to simulate the effect of different sampling rates on the observed measure of inequality, drawing from a population at a single point in time so that there is no change in actual income segregation. We find considerable variation in estimates across samples taken from the same population, particularly for smaller samples. The difference between the median estimate using sampling rates comparable to Census 2000 and the ACS is as large as the observed changes since 2000. We propose alternative approaches to calculate unbiased estimates of class segregation.