A quantitative, semi-automated method for classifying human sperm based on objective measurements of head shapes and sizes has been developed. Air-dried smears of semen from eight healthy men were stained with the Feulgen reaction and 283 sperm were selected as prototypic examples of the 10 morphology classes used in our classification system. Sperm heads were imaged through a microscope (NA = 1.3), sampled at 0.125-micron intervals, and measured on an image analysis system. Measurements included stain content, length, width, perimeter, area, and arithmetically derived combinations. Additionally, each sperm image was optically sectioned at right angles to its major axis to give a measure of lengthwise heterogeneity of shape. Linear stepwise discriminant analysis was used to identify the more powerful parameters and to create a model employing eight parameters. The jackknifed classification procedure distinguished normal from abnormal sperm with 95% accuracy and correctly assigned 86% of the sperm to one of 10 shape classes. Most of the misclassification errors occurred among closely related classes. The results demonstrate the ability of automated image analysis to classify individual sperm into clinically familiar shape categories.