Artificial selection experiments are potentially powerful, yet under-utilized tool of evolutionary and physiological ecology. Here we analyze and review three important aspects of such experiments. First, we consider the effects of instrumental measurement errors and random fluctuations of body mass on the total phenotypic variation. We illustrate this with the analysis of measurements of oxygen consumption in an open-flow respirometry set-ups. We conclude that measurement errors and fluctuations of body mass are likely to reduce the repeatability of oxygen consumption by about one third. Using published estimates of repeatability of metabolic rates we also showed that it does not tend to decline with increasing time between measurements. Second, we review data on narrow sense heritability (h(2)) of metabolic rates in mammals. The results are equivocal: many studies report very low (∼0.1) h(2), whereas some recent studies (including our own estimates of h(2) in laboratory mice, obtained by means of parent-offspring regression) report significant h(2) ≥ 0.4. Finally, we discuss consequences of the lack of replicated lines in artificial selection experiments. We focus on the confounding effect of genetic drift on statistical inferences related to primary (selected) and secondary (correlated) traits, in the absence of replications. We review literature data and analyze them following the guidelines formulated by Henderson (1989, 1997). We conclude that most results obtained in unreplicated experiments are probably robust enough to ascribe them to the effect of selection, rather than genetic drift. However, Henderson's guidelines by no means should be treated as a legitimate substitute of the analysis of variance, based on replicated lines.