The aim of this study was to assess the accuracy of GPS 3300 to predict active (feeding, moving) and inactive (bedded, standing) behaviours in European roe deer. Observed activity of three tame roe deer was correlated with the three variables provided by the activity sensors of their collars (normalised for each individual) using discriminant analyses. The best model discriminated correctly 84% of the "active" time intervals (defined as roe deer spending at least 30% of the time active) and 97% of inactive ones. We tested for consistency with movements using 11 wild roe deer, six living in a forest and five in an open habitat. As expected, wild deer identified as inactive (sensors) were mostly immobile (GPS locations, 831 records of 1096), whereas deer known to be moving were mostly classified as active (493 of 758 records). The records classified as inactive but moving (24% of inactive records or 35% of moving records) can indicate short distance displacement, erroneous detection of movement, or moving without neck movements. The records classified as active but immobile (46% of active ones, 33% of immobile ones) can represent feeding on a small patch without detectable movements.