Several predictors of cognitive impairment assessed by Mini Mental State Examination (MMSE) have previously been identified. However, which predictors are the most relevant and what is their effect on MMSE categories remains unclear. Cross-sectional and longitudinal study using data from 1116 older adults (72.6 ± 5.6 years, 579 female), 350 of whom were followed for 7 years. At baseline, the following variables were collected: personal data, marital status, occupation, anthropometric measures, risk factors, previous cardiovascular events, self-rated health and physical activity during the last week. Furthermore, routine laboratory tests, abdominal echography and a step test (with measurement of the time needed to ascend and descend two steps 20 times) were performed. The associations of these variables with cross-sectional cognitive deficit (MMSE < 24) and longitudinal cognitive decline (decrease of MMSE score over 7 years of follow-up) were investigated using logistic regression models. Cross-sectional cognitive deficit was independently associated with school education ≤ 5 years, prolonged step test duration, having been blue collar or housewife (P ≤ 0.0001 for all) and, with lower significance, with advanced age, previous stroke and poor recent physical activity (P < 0.05). Longitudinal cognitive decline was mainly associated with step test duration (P = 0.0001) and diastolic blood pressure (P = 0.0002). The MMSE categories mostly associated with step test duration were orientation, attention, calculation and language, while memory appeared to be poorly or not affected. In our cohort of older adults, step test duration was the most relevant predictor of cognitive impairment.