Abstract.-I used phenotypic selection analysis to test the prediction from functional and comparative studies of plants that smaller leaves and more efficient water use are adaptive in drier environments. I measured selection gradients on leaf size and instantaneous water-use efficiency (a measure of carbon gain per unit water loss) in experimental populations of Cakile edentula var. lacustris placed into wet and dry environments in the field. Linear and nonlinear selection differed significantly between the two environments as predicted. Water-use efficiency was selected to be higher, and leaf area was selected toward a small intermediate optimum, in the dry environment. There was also significant positive correlational selection on water-use efficiency and leaf size, suggesting that the optimum leaf size in the dry environment is greater for plants with higher water-use efficiency. In contrast, neither leaf size nor wateruse efficiency were selected in the wet environment, though larger leaves resulted in greater vegetative biomass. Path analysis of the linear selection gradients found that water-use efficiency affected plant fitness primarily because it increased vegetative biomass, as suggested by the hypotheses about the function of physiological traits. These results were not only consistent with the functional hypotheses but also with the observed genetic differentiation in wateruse efficiency and leaf size between wet and dry site populations.Key words.-Adaptive phenotypic plasticity, Cakile, leaf size, multivariate selection analysis, path analysis, water-use efficiency.Received August 30, 1994. Accepted April 25, 1995 The evolutionary response to selection can be predicted by the phenotypic selection in one generation and the genetic variances and covariances (Lande and Arnold 1983). Phenotypic selection studies, which describe the relationship between traits and fitness, can be used not only to predict the evolution of the traits studied but also as an empirical test of adaptive hypotheses. To support the hypothesis that a trait is adaptive in a given environment, demonstrating that the trait is correlated with fitness in that environment is not sufficient. The correlation of trait and fitness must also be lessened or absent in an environment where the trait is not expected to be adaptive (Wade and Kalisz 1990). Many studies have demonstrated strong natural selection in the wild (Endler 1986), but few studies have attempted to test evolutionary predictions and adaptive hypotheses. Here I describe a test of adaptive hypotheses for plant physiological traits that affect plant carbon uptake and water loss. A large body of functional analyses and comparative studies provides hypotheses on how selection on these traits should depend on environmental water availability (i.e., Ehleringer 1975;Givnish 1986, Cowan 1986. In this paper, I describe a field study of selection on leaf size and water-use efficiency in wet and dry environments and compare the results with the predictions of adaptive hypotheses. In a ...