Measuring yield accurately is critical for evaluating the impact of interventions that aim to increase agricultural productivity but presents challenges in the case of coffee due to the long harvest period. An allometric approach, in which the fruits on randomly selected branches and clusters are counted is widely used due to its non-destructive nature and acceptability to farmers. However, this approach requires careful attention to detail, which may be difficult to maintain in the context of large-scale data collection efforts. Using data from 199 small-scale Robusta coffee farms in Uganda, we compare yield estimates obtained through a standard allometric protocol against those from a one-time harvest of both ripe and unripe cherries prior to the start of the harvest season. The one-time harvest method was widely acceptable to farmers. Allometric yield estimates explain just under half of the variation in the harvest-based yield measure. While estimated yield is similar across methods for the first tree harvested per farm, we observe a larger difference in allometric versus harvest-based estimates, and systematically lower counts of stems and branches for trees assessed later during the farm visit. We interpret these findings as evidence of deteriorating enumerator performance on the allometric method over time, implying a risk of downward-biased yield estimates.