A method of estimating doses from multiple dilutions of an unknown sample in radioimmunoassay is described. It uses a computer program to minimise the root mean square error about a standard curve. The confidence limits of the estimates were evaluated from the sum of squares error as a function of dose. An attached subroutine reported and suppressed points in the data set that showed unexpectedly large errors of fit. The program agreed well with an alternative point-by-point evaluation for most practical data but also responded robustly to errors at the limits of the assay curve. By studying patterns of point rejection, it is possible to build up information about the relation between the standard curve parameters and those in the unknown sets. The method is illustrated by studies of laboratory-based materials and by an assay of alphafetoprotein in samples of body fluids.