Using panel data from 2009, 2011, and 2015, this study estimates the impact of rice production training conducted in Uganda on the adoption of improved cultivation practices and productivities. Since participants were encouraged to share information with fellow farmers, the average effects on training participants and non-participants in training villages (spillover effects) are separately estimated by selecting comparable households from villages without training projects. Because of the non-random assignment of project villages and training participation, a difference-in-differences model with household fixed effects is combined with propensity score weighting for mitigating biases. We find that training increases adoption rates for improved cultivation practices among training participants, both in the short and long term, and the long-term impact of training on rice yield is 0.47 tons per hectare. Although non-participants in training villages increased the adoption of transplanting in the long term, no improvements in non-participants’ knowledge on rice cultivation nor in rice productivity were detected. The results of the heterogeneous impacts on non-participants’ adoption show non-participants who visited the demonstration plot increased the adoption of transplanting, but those who talked with training participants about rice cultivation did not increase the adoption rate more than those who did not.