The aim of this study was to improve the accuracy of the histopathologic diagnosis in the differential diagnosis between obstructive and nonobstructive forms of neonatal cholestasis, using this clinical situation as a model for a mathematical approach. The study was blind, and we performed it in two steps. In the first step, 49 histologic parameters were visually estimated and were scored on a scale of 0 to 4+ in 100 liver biopsy specimens obtained between 1980 and 1985 from 78 patients with neonatal cholestasis. Forty-eight of these 100 specimens were from patients with final diagnosis of obstructive cholestasis (Group I), and 52 were from patients with nonobstructive cholestasis (Group II). The age range was 3 to 24 weeks (median, 12.5 wk). Twelve histologic variables were selected by chi 2 and Fisher's exact test (P < .05). Next, a series of combinations among these variables were submitted to statistical analysis by logistic regression method, defining a six-variable model that had the most powerful predictive value to classify the type of cholestasis. The variables were portal ductal proliferation, bile plugs in portal bile ductules, portoportal bridges, neutrophils, hepatocyte swelling, and multinucleated giant hepatocytes. The score obtained by this model correspond to the probability of a case belonging to Group I. The accuracy, sensitivity, and specificity rates were 94.0%. In the second step, the model was applied to a new sample of 74 needle-liver biopsy specimens obtained between 1990 and 1995, 45 from patients in Group I and 29 from patients in Group II. The age range was 3 to 15 weeks (median, 8 wk). The accuracy, sensitivity, and specificity rates were 90.5%, 100%, and 75.9%, respectively. In our diagnostic routine, this score has been systematically reported and has been helpful in orienting the therapeutic decision in this group of patients.