Fluctuation analysis is an established and widely used technique of estimating mutation rates in cultured cells. This paper presents a modified median estimator of mutation rates, which is novel because it allows for unequal population sizes N(t) of the parallel cultures, and helps to detect and reduce the estimation variability. Simulation results show a good accuracy and robustness of the modified median estimator compared with the median estimator and the maximum likelihood estimator. The proposed estimator, based on the Luria-Delbrück model, is applied to 20 yeast datasets collected during 3 different days for a study of chromosome loss and recombination in wild-type Saccharomyces cerevisiae strains. The estimates obtained display among-experiment variability, which is inflated with respect to the model predictions on simulated data. Further investigation in S. cerevisiae and Escherichia coli uncovers an empirical inverse relationship between the population sizes N(t) and the mutation rate estimates under certain experimental conditions. The impact of these effects on the practice of fluctuation analysis is discussed.