Objective To determine the relationship of total and free serum testosterone to cognitive performance in older men.Design Cross-sectional study of a population-based sample.Participants A total of 2932 men aged 70-89 years.Measurements Cognitive function was assessed using the Standardized Mini-Mental State Examination (SMMSE). Early morning sera were assayed for total testosterone, SHBG and LH. Free testosterone was calculated using the Vermeulen method.Results There were weak positive correlations between SMMSE score and serum free testosterone (Spearman's rho = 0.06, P = 0.001) and total testosterone (r = 0.04, P = 0.027), and a weak negative correlation with LH (r = -0.07, P < 0.001). Men with SMMSE scores in the top quintile had higher serum free testosterone compared with those in the lowest quintile [median (interquartile range, IQR): 278 (228-335) vs. 262 (212-320) pmol/l, P = 0.003], but similar total testosterone [15.2 (11.9-18.8) vs. 14.8 (11.6-18.3) nmol/l, P = 0.118]. Increasing age, non-English-speaking background, lower educational attainment, presence of clinically significant depressive symptoms, and cardiovascular morbidity were associated with the lowest cognitive performance quintile. After their effects were taken into account in a multivariate analysis, serum free testosterone >= 210 pmol/l was associated with reduced likelihood of poor cognitive performance on the SMMSE [odds ratio (OR) 0.71, 95% confidence interval (CI) 0.52-0.97].Conclusions In community-dwelling older men, serum free testosterone >= 210 pmol/l is associated with better cognitive performance. In this context, calculated free testosterone seems to be a more informative measure of androgen status than total testosterone. Studies examining the contribution of androgens to age-related cognitive decline should incorporate an assessment of free testosterone concentration.