Natural selection acts on phenotypes, and seldom operates on single phenotypic traits and correlations between characters seriously complicate the measurement of selection. Linear and nonlinear selection gradients can be used to infer selection from phenotypic data and quantify and visualize how traits affect fitness, and fitness surface can show selection patterns clearly. I wonder why this tool is not extensively used (admitting my ignorance for it) even if I recognize that a good background in mathematics and statistics is required.