A complete understanding of the role of natural selection in driving evolutionary change requires accurate estimates of the strength of selection acting in the wild. Accordingly, several approaches using a variety of data-including patterns of DNA variability, spatial and temporal changes in allele frequencies, and fitness estimates-have been developed to identify and quantify selection on both genotypes and phenotypes. Here, we review these approaches, drawing on both recent and classic examples to illustrate their utility and limitations. We then argue that by combining estimates of selection at multiple levels-from individual mutations to phenotypes-and at multiple timescales-from ecological to evolutionary-with experiments that demonstrate why traits are under selection, we can gain a much more complete picture of the adaptive process.