This paper demonstrates that fetal heart rate (FHR) patterns can be classified by algorithmically determined linear discriminants. A nonparametric learning algorithm was applied to 17 samples of five-vectors. The coordinates of each sample vector were visual features derived from the FHR curve and the simultaneous uterine contraction pressure data in accord with medical training-literature. Data were obtained from strip-chart recordings from the Cedars-Sinai Medical Center, Los Angeles, where an FHR monitoring and on-line computer processing system based on an IBM System/7 is being installed. The algorithm converged to linear discriminants that correctly classified all the 17 training samples under four different combinations of initial weights, training sequence, and correction increment. Each of the four linear decision rules so obtained was applied to 14 new sample vectors. Three classified 11 samples correctly and one classified 13 samples correctly. Medical anomalies (atypical data) were present in all three misclassified patterns. A perfect success record was found in classifying all seven medically ominous new sample vectors.