In several reviews, exercise was reported to be effective in reducing the risk for cognitive decline and dementia [1,2]. However, not all reviews concluded this. One recent review [3] stated that there was still insufficient evidence, as most studies were too small and had insufficient methodological information (intensity, duration) to enable appropriate evaluation. Here we discuss potential confounds or mediators that may explain these discrepancies.We found that most observational studies showed at least some positive associations of exercise, but not always on the same cognitive tests [1]. Variance in studies was induced by inconsistent use of cognitive assessments (e.g. fluid intelligence compound scores vs Symbol Digit Modalities Test (SDMT) scores by itself); different assessments of fitness (objective vs. self reported hours of exercise engaged in); and different cut-offs for high/low exercise across studies. These limitations were echoed by others and recent reviews also illuminated potential confounds associated with both exercise and cognitive improvement, such as lifting of depression, as well as social and cognitive stimulation [4], and an increase in self efficacy (Stock, in press), which have usually not been taken along in analyses or adequately controlled for. Observational studies are limited in their ability to establish causality and many people could have stopped exercise because of other confounding morbidity, which may also affect cognitive function (e.g. vascular disease, see below). Randomised controlled trials (RCT) are better at establishing causality, but can also be affected by choice of measurements and population, suffer from baseline differences, regression to the mean and design of the control conditions (e.g. without social or cognitive stimulating aspects), as well as the above mentioned limitations of potential non-assessed confounds or mediators, such as mood.In our earlier review of 26 RCT studies in community dwelling elderly without known dementia or cognitive impairment, which had been carried out up to 2009 [1], only 6 studies showed overall cognitive improvement, 13 some improvement and 7 none at all. The most consistent cognitive tests to be affected by exercise interventions in this group were simple tests, such as those of concentration and those using simple reaction times. Several earlier reviews suggested that more complex cognitive tests were most affected by exercise. However, our review finding was substantiated by those of the Cochrane meta analyses [5], which is a gold standard medical review system. About half of RCT studies we had included in our review [1], which had used a simple test of concentration and working memory (Digit Span) found that it displayed significant positive results of exercise, but the other half of studies using this test (n=5) had not found any improvement. Of the complex information processing tests previously thought to be most sensitive to exercise, 21 tests (including the Stroop 4x, Symbol Digit Modalities test (SDMT) 4x, CRT 3x,...