Errors and error sources occurring in rotating-analyzer ellipsometry are discussed. From general considerations it is shown that a rotating-analyzer ellipsometer is inaccurate if applied at P = 00 and in cases when 0 = 0' or where A is near 00 or 1800. Window errors, component imperfections, azimuth errors and all other errors may, to first order, be treated independently and can subsequently be added. Explicit first-order expressions for the errors 6A and 36t' caused by windows, component imperfections, and azimuth errors are derived, showing that all of them, except the window errors, are eliminated in a two-zone measurement.Higher-order errors that are due to azimuth errors are studied numerically, revealing that they are in general less than 0.10. Statistical errors are also discussed. Errors caused by noise and by correlated perturbations, i.e., periodic fluctuations of the light source, are also considered. Such periodic perturbations do cause random errors, especially when they have frequencies near 2 COA and 4 C0A.