Population abundance data vary widely in quality and are rarely accurate. The two main components of error in such data are observation and process error. We used Bayesian state space models to estimate the observation and process error in time-series of 55 globally distributed populations of two species, Cervus elaphus (elk/red deer) and Rangifer tarandus (caribou/reindeer). We examined variation among populations and species in the magnitude of estimates of error components and density dependence using generalized linear models. Process error exceeded observation error in 75% of all populations, and on average, both components of error were greater in Rangifer than in Cervus populations. Observation error differed significantly across the different observation methods, and predation and time-series length differentially affected the error components. Comparing the Bayesian model results to traditional models that do not separate error components revealed the potential for misleading inferences about sources of variation in population dynamics.