Age-related white matter lesions (WMLs) are often subdivided into periventricular and subcortical WMLs. We previously reported that periventricular lesions seem more relevant for cognitive decline and dementia. However, whether this differentiation is relevant remains debated. The interpretation of studies on the discrepancies in aetiology and clinical course is impeded by the use of different visual classification methods with limited reproducibility. We developed a fully automated method to classify periventricular and subcortical WMLs and validated this against the results from a visual rating. The automated classification method was developed and tested on 490 persons from the Rotterdam Scan Study, a population-based cohort study on diseases among the elderly. For each subject 3 different MR sequences, a WML segmentation obtained with an automated method, and a semi-quantitative lesion classification by two trained observers were available. For the automated WML classification, the lesions were classified based on their three-dimensional distance to the ventricles, which were segmented using an automated method. Maximizing the Pearson correlation coefficient between the automatically classified WML volumes and the rating assigned by the observers on a random subset of 100 subjects, suggested a distance of 7mm as the optimal cut-off distance to distinguish periventricular from subcortical WMLs. This was validated in the remaining 390 persons by computing correlations between the automated and observer-based lesion scores. Subsequently, we assessed whether the automated classification replicated previous findings of a different effect of periventricular and subcortical lesion loads on the risk of dementia. Correlation coefficients between automated and visual ratings were 0.71 (p-value<0.001) for periventricular WMLs and 0.60 (p-value<0.001) for subcortical WMLs. The hazard ratios of dementia per standard deviation increase, adjusted for sex and age, were 1.87 (95%-CI: 1.39-2.52) for periventricular lesions and 1.45 (95%-CI: 1.13-1.86) for subcortical lesions. Adjusted for each other, the hazard ratio slightly increased for periventricular WMLs (1.98 (95%-CI: 1.26-3.13)), but attenuated for subcortical WMLs (0.93 (95%-CI: 0.60-1.44)). The automated WML classification is comparable to that of trained observers. The results support the notion that periventricular and subcortical WMLs bear different risks on the development of dementia.