OBJECTIVE Physical activity (PA) and cognition have reportedly been associated with each other. However, it remains to be elucidated what intensities of PA are most strongly associated with cognition. In the current study, we aimed to determine the association between the intensities of objectively measured PA and cognitive function. METHODS The present study was a cross-sectional analysis of the data obtained at registration in a randomized control trial in Toyota, Japan. Participants were community-dwelling older adults who had cognitive complaints. A battery of neuropsychological and physical assessments was carried out. Daily PA data were collected with the activity monitor. PA was categorized into one of three activity levels defined as light (<3.0 metabolic equivalents; LPA), moderate and vigorous (3.0 metabolic equivalents) activity. Partial correlation analysis was used to investigate the correlation between PA and cognition, with adjustments for age, sex and school years. We then carried out a multiple regression analysis to investigate the association of cognitive performance with PA, adjusting for insulin resistance or depressive mood. RESULTS Partial correlation adjusted for age, sex and schooling years showed that LPA was significantly correlated with the Digit Symbol Substitution test, Trail Making Test (TMT) part A and TMT-B, whereas moderate and vigorous activity showed no correlations. Multiple regression analysis with several models with different adjustments showed that LPA was associated with the Digit Symbol Substitution test, TMT-A and TMT-B independently from insulin resistance or depressive mood. CONCLUSIONS In the current study, we found that LPA was significantly associated with the performance of executive functional assessments. Geriatr Gerontol Int 2018; 18: 922-928.