In forested ecosystems, estimating the abundance or trend of most wildlife populations is difficult. Therefore, vital rates are often used to model population change, but validating such models is important. Using data from woodland caribou (Rangifer tarandus), we compared estimates of population change (λ) based on vital rate models to λ based on aerial censuses. We modeled λ using Hatter and Bergerud's (1991) recruitment‐mortality (R‐M) equation (λ = survival/[1 − recruitment]). We estimated survival and recruitment from a sample of 317 radio‐collared caribou from 9 subpopulations in British Columbia, Canada. In this ecosystem, woodland caribou have high sightability (>85%) in winter and thus are easy to census compared to most forest wildlife. We found that the R‐M equation overestimated λ compared to census‐based λ across most of the observed range of data (e.g., if R‐M estimated λ of 1.1, census‐based λ was 0.99, and if R‐M was 0.90, census‐based λ was 0.89). We then assessed whether recruitment, survival, a linear model of both parameters, or the R‐M equation best predicted census‐based λ. The R‐M equation explained 60% of the variation in census‐based λ, more than double the next‐best approach (i.e., the simple linear model), even though identical parameters were included. Further, we simulated variability due to the unknown sex (M:F) ratio in the sample, and found that the R‐M equation remained the best predictor of census‐based λ. Although the R‐M equation was the most precise and accurate approach, our results reaffirm that it is important to periodically validate trend estimates based on vital rate models with estimates of absolute abundance, particularly for species of management concern. © 2016 The Wildlife Society.