Rice is an important cereal to Ghana’s economy and agriculture. While there is huge potential for lowland rice cultivation in northern Ghana, erratic rainfall and low and degrading soil fertility remain the key constraints of production. The present study analyzes the adoption decision and the impact of bund technology and dibbling method on fertilizer demand, output supply, and net returns. The present analysis explicitly takes account of selection bias by employing endogenous switching regression and propensity score matching. The cross-sectional data set refers to the cropping season 2005 and consists of 342 smallholder rice farmers. Results of the seemingly unrelated bivariate probit model suggest that the adoption decisions of dibbling and bund technology should be estimated jointly. Estimates indicate that bunds are more likely constructed on marginal land as a preventive technology, while dibbling method seems to be used complementary to good soil productivity. Adoption decisions are found to be related to economic constraints, particularly of labour and capital, while off-farm income appears to decrease the adoption of labour-intensive technologies. Furthermore, adoption seems to be strongly related to the use of interrelated technologies, the perception of technologies, participation in technology-related projects and farmer groups, plot-level characteristics, and geographic location. To examine the impact of technology adoption, this study applies Mahalanobis metric matching with calipers and the propensity score as additional variable. This approach is particularly useful in the present analysis with multiple treatments. To check for the robustness of results, kernel based matching and nearest neighbour matching are applied. Balancing tests were conducted by checking the reduction of the mean standardized absolute bias. To control for hidden bias due to selection on unobservables, sensitivity analysis was done by employing the Rosenbaum (2002) bounding approach. Results indicate the significance of matching in reducing bias in the distribution of relevant variables between the treatment and control group and that the estimates are quite insensitive to hidden bias. Results of the Mahalanobis metric matching indicate that the adoption of bund technology has a positive and significant effect on fertilizer demand, as well as a positive, but insignificant impact on output supply and net returns. Adopters of dibbling appear to have higher rice yields, while no significant difference in net returns and fertilizer demand was found. However, data reveal a positive and significant effect on output supply and net returns when dibbling method is combined with intensified weeding. Furthermore, when dibbling is not only used as seed sowing but also as fertilizer application method, nitrogen demand is significantly higher. The estimates of the endogenous switching regression model suggest that self-selection occurs. Labour and capital constraints, the use of interrelated technologies, social networks such as farmer groups, education, learning effects through the dissemination and the use of interrelated technologies, and the timely availability of land preparation equipment appear to be important factors in determining farm outcomes. However, the effects vary in the level and significance according to the type of technology and outcome. This supports the need to examine the impact of determinants separately for adopters and non-adopters.