The evidence for nonnormality of crop yields is reassessed. Three methodolo
gical problems are identified in typical yield distribution analyses: (i) m
isspecification of the nonrandom components of yield distributions, (ii) mi
ssreporting of statistical significance, and (iii) use of aggregate time-se
ries (ATS) data to represent farm-level yield distributions. One or more of
these problems infect virtually all evidence against normality to date. Th
e positive contribution of the article is a set of principles that must be
followed in any valid investigation of yield normality.