regions than others. The centroid of the structurefactor probability distribution is obtained, in this case, by taking the Fourier transform of the expected electron-density function. In other terms, each atom of a molecular replacement model would be smeared over its distribution of possible positions. It is assumed that there is a sufficient number of independent contributions to the difference in the structure factors, so that the central limit theorem applies and the probability distribution is a Gaussian about the centroid estimate.For a model of a crystal structure, it is preferable to consider the average effect of a specific set of errors on a set of structure factors, in other words to consider the reciprocal-space vector as the random variable. The probability distributions underlying the differences between the model and the true structure enter through the frequencies of the errors over all the atoms. Essentially the same probability distributions of structure factors arise as in the previous case, because of the symmetry between real and reciprocal space in the Fourier transform.Considered in terms of normalized structure factors, all sources of error have the same effect, which can be summarized in a single parameter, o¥. This parameter plays the same role in the probability distributions as O" A in the distributions of Srinivasan & Ramachandran (1965). Therefore, the methods suggested previously to estimate phase probabilities and to calculate electron-density maps (Read, 1986) are still valid. However, the interpretation of the parameter ere is different. In particular, the variation of o¥ with resolution cannot be attributed entirely to coordinate error. Methods such as the Luzzati (1952) plot and the o'a plot (Read, 1986) (1987). Protein Eng. 1,377-384. WILSON, A. J. C. (1949). Acta Cryst. 2, 318-321. WILSON, A. J. C. (1976). Acta Cryst. A32, 781-783. WOOLFSON, M. M. (1956). Acta Cryst. 9, 804-810. Acta Cryst. (1990