We investigated the effects of varying census duration on line transect estimates of winter bird abundance and species richness in second-growth coastal western hemlock forest fragments in Vancouver, British Columbia. Three treatments of census duration (12 min, 18 min and 24 min) were compared on 200 m-long transects at three study sites over three fortnights in a replicated latin square experimental design. Overall, census duration affected the number of individual birds and bird species detected. The number of individual birds detected increased significantly with census duration from 12 min to 18 min but not from 18 min to 24 min. The number of bird species detected also increased with census duration, although the increase was significant only between 12 min and 24 min. The results suggest that for winter bird counts in coastal western hemlock forest, a census duration of 18 min for a 200 m-long transect is most efficient.