Limited data on toxic compounds in humans often necessitate that biological half-life estimates be based on only two measurements over a time interval, both of which are subject to analytical error. Propagation of error calculations and computer simulations were performed which showed that as the time interval gets shorter and/or the analytical error increases, the variability of the half-life estimate increases. Negative and large positive values create problems in computing the mean half-life and can lead to biases in that estimator. However, when negative values are considered equivalent to large positive half-lives, the median half-life has negligible bias even at large analytical coefficients of variation and measurement intervals much shorter than one half-life.