This paper discusses the statistical and biological issues involved in using Lande and Arnold's (1983) multiple-regression approach to measure natural selection. The method involves measuring relative fitness and phenotypic characters in a population to estimate the effects of these characters on fitness. While this approach allows for direct assessment of selection, it has limitations, such as multicollinearity and unmeasured factors affecting fitness, which can complicate inference. The paper also highlights the importance of statistical assumptions, such as normality of residuals, and suggests using jackknife tests for significance. It emphasizes that while fitness regression can suggest hypotheses about selection, manipulative experiments are needed to confirm causal relationships. The paper also discusses alternative definitions of stabilizing and disruptive selection. The authors caution that the method may not always be reliable due to violations of statistical assumptions and suggest that alternative approaches, such as discriminant analysis and logistic regression, may be more appropriate in some cases. The paper concludes that while regression analysis is a valuable tool for predicting evolutionary responses, it should be used with caution and in conjunction with other methods to ensure accurate interpretation of results.This paper discusses the statistical and biological issues involved in using Lande and Arnold's (1983) multiple-regression approach to measure natural selection. The method involves measuring relative fitness and phenotypic characters in a population to estimate the effects of these characters on fitness. While this approach allows for direct assessment of selection, it has limitations, such as multicollinearity and unmeasured factors affecting fitness, which can complicate inference. The paper also highlights the importance of statistical assumptions, such as normality of residuals, and suggests using jackknife tests for significance. It emphasizes that while fitness regression can suggest hypotheses about selection, manipulative experiments are needed to confirm causal relationships. The paper also discusses alternative definitions of stabilizing and disruptive selection. The authors caution that the method may not always be reliable due to violations of statistical assumptions and suggest that alternative approaches, such as discriminant analysis and logistic regression, may be more appropriate in some cases. The paper concludes that while regression analysis is a valuable tool for predicting evolutionary responses, it should be used with caution and in conjunction with other methods to ensure accurate interpretation of results.