1. Outdoor recreation on trail networks is a growing form of disturbance for wildlife. However, few studies have examined behavioural responses by large carnivores to motorised and non-motorised recreational activity-- a knowledge gap that has implications for the success of human access management aimed at improving habitat quality for wildlife. 2. We used an integrated step-selection analysis of grizzly bear (Ursus arctos) radiotelemetry data and a spatio-temporal model of motorised and non-motorised human recreational activity to examine the effect of human recreational activity along trails on both habitat selection and movement behaviour of individual bears. Grizzly bears were captured and radiocollared in the west-central Alberta Rocky Mountains and Foothills, and trail cameras were deployed on trails to obtain data on human recreational activity. 3. We found that models including data on recreational activity outperformed trail-proximity models when interactions with movement covariates were included. Responses were highly variable among individuals, and across classes; males, females and females with cubs. 4. Male and solitary female grizzly bears increased avoidance of trails with a high probability of motorised activity, as well as displaying increased movement rates in response to motorised recreation. Females with cubs did not increase avoidance, however they had the largest response with higher movement rates. In contrast, for all classes selection for proximity to trail increased when probability of non-motorised activity was high, and the effect on movement was dampened relative to the motorised response. 5. Synthesis and applications. By combining selection and movement into a unified modelling framework, we show that bears alter selection and movement behaviour in response to trails and recreation, and that such responses are determined by the type of recreational activity. Reduced selection and increased movement in proximity to motorised trails could affect bears’ ability to exploit foraging opportunities in these areas. Future access management actions for grizzly bear recovery should consider frequency and type of linear feature use by humans rather than solely relying on thresholds relating to feature densities.