Previous studies have shown that terrestrial lidar is capable of characterising forest canopies but suggest that lidar underestimates gap fraction compared to hemispherical camera photography. This paper performs a detailed comparison of lidar to camera-derived gap fractions over a range of forest structures (in snow affected areas) and reasons for any disagreements are analysed. A terrestrial laser scanner (Leica C10 first return system) was taken to Abisko in Northern Sweden (sparse birch forests) and Sodankylä in Finland (spruce and pine forests) where five plots of varying density were scanned at each (though one Abisko plot was rejected due to geolocation issues). Traditional hemispherical photographs were taken and gap fraction estimates compared. It is concluded that, for the sites tested, the reported underestimates in gap fraction can be removed by taking partial hits into account using the return intensity. The scan density used (5–8 scans per 20 m by 20 m plot) was sufficient to ensure that occlusion of the laser beam was not significant. The choice of sampling density of the lidar data is important, but over a certain sampling density the gap fraction estimates become insensitive to further change. The lidar gap fractions altered by around 3–8% when all subjective parameters were adjusted over their complete range. The choice of manual threshold for the hemispherical photographs is found to have a large effect (up to 17% range in gap fraction between three operators). Therefore we propose that, as long as a site has been covered by sufficient scan positions and the data sampled at high enough resolution, the lidar gap fraction estimates are more stable than those derived from a camera and avoid issues with variable illumination. In addition the lidar allows the determination of gap fraction at every point within a plot rather than just where hemispherical photographs were taken, giving a much fuller picture of the canopy. The relative difference between TLS (taking intensity into account) and camera derived gap fraction was 0.7% for Abisko and −2.8% for Sodankylä with relative root mean square errors (RMSEs) of 6.9% and 9.8% respectively, less than the variation within TLS and camera estimates and so bias has been removed.