Aims/hypothesis The study aimed to investigate whether baseline physical activity protects against the occurrence of type 2 diabetes during a 28 year follow-up, after controlling for childhood environment and genetic predisposition. Methods At baseline in 1975 same-sex twin pairs born in Finland before 1958 were sent a questionnaire including questions on physical activity. The participants (20,487 individuals, including 8,182 complete twin pairs) were divided into quintiles by leisure-time physical activity metabolic equivalent (MET) index (MET h/day). Type 2 diabetes was determined from nationwide registers for the follow-up period (1 January 1976-31 December 2004). Individual and pairwise Cox proportional hazard models were used. Results During follow-up, 1,082 type 2 diabetes cases were observed. Among all individuals, participants in MET quintiles (Q) III-V had significantly decreased risk for type 2 diabetes compared with sedentary individuals (QI). The pairwise analysis on pairs discordant for physical activity showed that participants in MET QII to V had significantly lower hazard ratios (0.61, 0.59, 0.61, 0.61) compared with sedentary participants. These findings from the pairwise analysis persisted after adjusting for BMI. In the pairwise analysis, the BMI-adjusted hazard ratio for type 2 diabetes was lower for physically active members of twin pairs (combined QII-V) than for inactive co-twins (HR 0.54; 95% CI 0.37-0.78). Similar results were obtained for both dizygotic and monozygotic pairs, as well as for the subgroup of twin pairs defined as free of co-morbidities in 1981 (HR 0.36;