Linear regression yields minimum-variance, unbiased estimates of the adjustable parameters, provided only that the analyzed data be unbiased and of finite variance. If further the data are normally distributed, then so will be the estimated parameters. But frequently data are transformed before fitting, and if the original data are normal, the transformed data may not be. In particular, inversion and logarithmic conversion yield biased, non-Gaussian distributions, so least-squares analysis of such data yields biased, nonnormally distributed parameters, even when the transformed data are properly weighted in accord with the transformation. Monte Carlo calculations are used to study the effects of such nonnormal data in cases of relevance to the analysis of equilibrium and kinetics data (exponential decay, binding constants, enzyme kinetics, fluorescence quenching, adsorption). Typically 10(5) equivalent data sets are processed to obtain precise information about the parameter biases and distributions. The biases generally persist in the limit of an infinite number of data values, which means that the estimators are not only biased but also inconsistent.