Active commuting offers the potential to increase physical activity among adults by being built into daily routines. Characteristics of the route to work may influence propensity to walk or cycle. Geographic information system (GIS) software is often used to explore this by modelling routes between home and work. However, if the validity of modelled routes depends on the mode of travel used, studies of environmental determinants of travel may be biased. We aimed to understand how well modelled routes reflect those actually taken, and what characteristics explain these differences. We compared modelled GIS shortest path routes with actual routes measured using QStarz BT-Q1000X Global Positioning System (GPS) devices in a free-living sample of adults working in Cambridge and using varying travel modes. Predictors of differences, according to length and percentage overlap, between the two route sets were assessed using multilevel regression models and concordance coefficients. The 276 trips, made by 51 participants, were on average 27% further than modelled routes, with an average geographical overlap of 39%. However, predictability of the route depended on travel mode. For route length, there was moderate-to-substantial agreement for journeys made on foot and by bicycle. Route overlap was lowest for trips made by car plus walk (22%). The magnitude of difference depended on other journey characteristics, including travelling via intermediate destinations, distance, and use of busy roads. In conclusion, GIS routes may be acceptable for distance estimation and to explore potential routes, particularly active commuting. However, GPS should be used to obtain accurate estimates of environmental contexts in which commuting behaviour actually occurs. Public health researchers should bear these considerations in mind when studying the geographical determinants and health implications of commuting behaviour, and when recommending policy changes to encourage active travel.