Little is known about the effect of shelter design on sheltering behavior in horses. This study investigates shelter use by Icelandic horses kept outdoors 24 hours a day during the winter in Denmark and whether shelter use and levels of fecal cortisol metabolites (FCMs) are affected by (1) the number of entrances (1 vs. 2) and (2) a partition inside the shelter. The effects of weather conditions on shelter use are also investigated. Thirty-two Icelandic horses participated in the study. The horses were pastured in 8 groups of 4 horses, and each group had access to a shelter (30 m2), which in the first study period (5 weeks, Dec-Jan) had either 1 or 2 entrances (n = 4 groups per treatment). In the second study period (5 weeks, Jan-Feb), all shelters had 2 entrances and half were equipped with a partition inside the shelter (n = 4 groups per treatment). Infrared cameras were placed inside all shelters for recording of shelter use. Feces were collected weekly during the last 3 weeks of each study period. We found that groups with 2 entrances to their shelter used the shelters significantly more than groups with only 1 entrance (% pictures with at least 1 horse inside, median [25;75%]: 2 entrances: 12.6 [7;20] vs. 1 entrance: 3.0 [2;4], P = 0.029). In addition, horses with 1 entrance had significantly increased FCM levels (ng/g, mean ± SE: 2 entrances: 6.8 ± 0.5 vs. 1 entrance: 10.0 ± 1.2, P = 0.019). The partitions did not affect shelter use or FCM levels. In both study periods, the shelters were used mainly at night (light vs. dark hours: P < 0.001), and daily average temperatures below zero degree Celsius increased shelter use. We conclude that entrance conditions are crucial to the use of shelters by Icelandic horses during winter.