Overwintering strategies are important for the survival of resident birds in temperate climates and among the most important are adjustments in roosting behaviour. In cavity roosting birds, previous studies have frequently used contact checks of man-made nest boxes to quantify roost-site occupancy. However, there is a concern that occupancy rate estimated by this method may be biased due to procedural disturbance. In this study, we quantified this potential bias by examining the winter time occupancy of 182 nest boxes in a floodplain forest in the Czech Republic. Nest boxes were checked three times a month from November to February 2007–2010 by three methods with decreasing level of potential disturbance. We obtained 1319 records of roosting birds of three species, with 94% being Great Tits, Parus major. We found a considerable decline in nest box occupancy throughout the winter when using the contact method (capture and handling of the bird), whilst occupancy rates remained constant when using the two non-contact methods (visual inspection of the opened nest box; the inspection by Infra red light mini camera passed through the entrance). The contact method was also associated with lower reuse rate of individual nest boxes. In conclusion, the commonly used direct night checks of nest boxes caused a disturbance to roosting birds and thus can lead to biased conclusions when studying winter time roosting behaviour in birds. More generally, this study demonstrates that using nest boxes may introduce bias in studies conducted during the non-breeding season, similarly as has been demonstrated for studies conducted in the breeding season.