Measurements of the genetic variation and covariation underlying quantitative traits are crucial to our understanding of current evolutionary change and the mechanisms causing this evolution. This fact has spurred a large number of studies estimating heritabilities and genetic correlations in a variety of organisms. Most of these studies have been done in laboratory or greenhouse settings, but it is not well known how accurately these measurements estimate genetic variance and covariance expressed in the field. We conducted a quantitative genetic half-sibling analysis on six floral traits in wild radish. Plants were grown from seed in the field and were exposed to natural environmental variation throughout their lives, including herbivory and intra- and interspecific competition. The estimates of heritabilities and the additive genetic variance-covariance matrix (G) obtained from this analysis were then compared to previous greenhouse estimates of the same floral traits from the same natural population. Heritabilities were much lower in the field for all traits, and this was due to both large increases in environmental variance and decreases in additive genetic variance. Additive genetic covariance expressed was also much lower in the field. These differences resulted in highly significant differences in the G matrix between the greenhouse and field environments using two complementary testing methods. Although the G matrices shared some principal components in common, they were not simply proportional to each other. Therefore, the greenhouse results did not accurately depict how the floral traits would respond to natural selection in the field.