Jatropha curcas L. is a non-food crop quoted as a promising natural feedstock for biodiesel production in tropical and subtropical regions. Although an efficient mechanism of drought avoidance through stomatal control of transpiration has been demonstrated in this species, low carbon assimilation and growth rates preclude any advantage of such strategy under water limitation. Two greenhouse experiments were conducted with the objective of investigating varietal differences in water use as the trade-off between carbon isotope composition (δ13C) in different tissues and water use efficiency, in young plants of J. curcas. The first experiment was a survey with seven provenances of J. curcas under non-limiting water availability. There were no significant differences among provenances for leaf gas exchange rates, growth and whole-plant transpiration (T). However, significant effects of provenance and tissue (stem bark or leaf) in δ13C were demonstrated. The provenances CNPAE183 and CNPAE222 were selected for the second experiment, as variations in water use traits and δ13C between the two provenances were observed. Soil water deficit, imposed for 18 days, led to significant physiological, biochemical, and morphological changes. An early and sharp response to water deficit in CNPAE183 as compared to CNPAE222 was observed, as indicated by a 44% higher rate of decrease of T in the former. In addition, water deficit led to increase of δ13C, which was more pronounced in CNPAE222 (13% in young leaves) when compared to CNPAE183 (11% in young leaves). In both provenances, δ13C was less negative in young as compared to mature tissues. Significant and direct correlations between stem bark and leaf δ13C and leaf-level intrinsic water use efficiency were observed. Contrasting results, particularly on T and δ13C, suggest the existence of genetic diversity for water relations and metabolic traits linked to drought tolerance in J. curcas. Using stem bark or leaf δ13C measurements as the basis for large-scale screening of water-use efficient genotypes should be considered.