Natural populations are composed of individuals that vary in their morphological traits, timing and interactions. The distribution of a trait can be described by several dimensions, or mathematical moments-mean, variance, skew and kurtosis. Shifts in the distribution of a trait across these moments in response to environmental variation can help to reveal which trait values are gained or lost, and consequently how trait filtering processes are altering populations. To examine the role and drivers of intraspecific variation within a trait filtering framework, we investigate variation in body size among five wild bumblebee species in the Colorado Rocky Mountains. First, we examine the relationships between environmental factors (climate and floral food resources) and body size distributions across bumblebee social castes to identify demographic responses to environmental variation. Next, we examine changes in the moments of trait distributions to reveal potential mechanisms behind intraspecific shifts in body size. Finally, we examine how intraspecific body size variation is related to diet breadth and phenology. We found that climate conditions have a strong effect on observed body size variation across all distributional moments, but the filtering mechanism varies by social caste. For example, with earlier spring snowmelt queens declined in mean size and became negatively skewed and more kurtotic. This suggests a skewed filter admitting a greater frequency of small individuals. With greater availability of floral food resources, queens increased in mean size, but workers and males decreased in size. Observed shifts in body size variation also correspond with variation in diet breadth and phenology. Populations with larger average body size were associated with more generalized foraging in workers of short-tongued species and increased specialization in longer-tongued workers. Altered phenological timing was associated with species- and caste-specific shifts in skew. Across an assemblage of wild bumblebees, we find complex patterns of trait variation that may not have been captured if we had simply considered mean and variance. The four-moment approach we employ here provides holistic insight into intraspecific trait variation, which may otherwise be overlooked and reveals potential underlying filtering processes driving such variation within populations.