The relationship of paternal age to specific types of trisomy and to chromosomally normal loss was investigated in data drawn from a case-control study of spontaneous abortions. Differences in paternal age between karyotype groups and controls delivering after 28 weeks gestation were tested using an urn model analysis which adjusted, by regression, for maternal age and, by stratification, for the effects of design variables (payment status, phase of study) and demographic factors (language, ethnicity). The magnitude of paternal age differences was estimated using least squares regression analysis. For chromosomally normal cases there was no association with paternal age. Among the fourteen trisomy categories examined, four (7, 9, 18, 21) showed increased paternal age (greater than or equal to 1 year above expectation), three (13, 20, 22) showed decreased paternal age and the rest, including the most common, trisomy 16, showed negligible differences. Only the association with trisomy 22 was statistically significant (P = 0.012), with a predicted reduction in paternal age of 2.1 years (95% CI -4.9, -0.5 years). This association did not vary with maternal age, payment status, phase of study, language or ethnicity. Because previous observations are extensive, the relation of paternal age to trisomy 21 was examined further. The overall association was not significant (beta = 0.8 years; 95% CI -0.8, 2.4 years). Moreover, there was evidence that the magnitude and direction of paternal age associations vary significantly within the sample, although not between subgroups defined on the basis of payment, phase of study, language or ethnicity. With respect to maternal age, the trend is towards a greater paternal age difference for trisomy 21 losses in younger women (P = 0.058). Given the number of tests performed, the finding for trisomy 22 and reduced paternal age could be due to chance. Among trisomy types, the direction of paternal age associations was not consistent for chromosomes grouped according to characteristics that might relate to the probability of nondisjunction, such as size, arm ratio, or nucleolar organizer region content, or to the potential viability of the trisomy. Thus, neither on statistical nor biological grounds do the data provide compelling evidence of paternal age effects on the trisomies found among spontaneous abortions, or on chromosomally normal losses.