The study of phenotypic variation in time and space is central to evolutionary biology. Modern geometric morphometrics is the leading family of methods for the quantitative analysis of biological forms. This set of techniques relies heavily on technological innovation for data acquisition, often in the form of 2D or 3D digital images, and on powerful multivariate statistical tools for their analysis. However, neither the most sophisticated device for computerized imaging nor the best statistical test can produce accurate, robust and reproducible results, if it is not based on really good samples and an appropriate use of the ‘measurements’ extracted from the data. Using examples mostly from my own work on mammal craniofacial variation and museum specimens, I will show how easy it is to forget these most basic assumptions, while focusing heavily on analytical and visualization methods, and much less on the data that generate potentially powerful analyses and visually appealing diagrams.