There have been competing arguments about the effect of public infrastructure on productivity in the literature. Level-based regressions generally show a much higher return to public capital than private capital, while difference-based regressions tend to find insignificant or even negative effects. To help reconcile this debate, this paper proposes that researchers should first test for causality in their data to check for length of lagged relationships and the existence of reverse causality, as a critical step before specifying a final model and estimating procedure on the relationship between the stock of capital and productivity growth. A newly developed system GMM method of estimation is proposed for this purpose. Second, a new method of estimating the relationship between the capital stock and productivity in level form is proposed that controls for possible endogeneity problems arising from reverse causation. These methods are illustrated using a unique set of pooled time-series, cross-section data for India. It is shown that infrastructure development in India is productive with an estimated impact lying between those obtained from level-based and difference-based estimates.