Despite the longstanding centrality of agricultural productivity growth as a development goal, reliable productivity measures remain elusive and costly. Crop yield is widely used as the primary productivity indicator, though it ignores inputs other than land and can be a poor proxy, particularly for smallholder and women farmers who are more likely to farm marginal lands and to grow multiple crops on the same plot, practices which complicate per hectare yield measurement. Crop yield is commonly calculated as a measure of production per harvested area, rather than production on the full area planted to the crop. Yet small-scale farmers are more likely to experience a loss in crop area between planting and harvesting, leading to systematic overestimation of mean crop yields so long as the null production on abandoned cropland goes unaccounted for. As a result, common yield measures may not be reliable indicators of aggregate agricultural productivity among smallholder farmers. We use plot-level data from the Tanzania National Panel Survey to investigate the conditions and crops for which the choice of yield measure might introduce significant error into crop yield estimates, and thereby bias research findings. We focus on three crops: maize, rice, and sorghum. We find that the choice of yield measure may lead to consistent under- or over-estimates of yield for sub-populations and crops that experience frequent and substantial losses in plot area between planting and harvest, with implications for the design of policy interventions to increase agricultural productivity and to target the least productive and poorest farmers.