A typical two-digit industry in the United States appears to have approximately constant returns to scale. Three puzzles emerge, however. First, estimates rise at higher levels of aggregation. Second, apparent decreasing returns contradicts evidence of small economic profits. Third, estimates with value added differ substantially from those with gross output. A representative-firm paradigm cannot explain these puzzles but a simple story of aggregation over heterogeneous units can. The authors discuss implications of heterogeneity for calibrating one-sector macroeconomic models, showing that these models sometimes require firm-level parameters but at other times require the 'biased' aggregate parameters. Copyright 1997 by the University of Chicago.