During the 1980s, all Japanese automobile producers opened assembly plants in North America. Industry analysts and previous research claim that these transplants are more productive than incumbent plants and that they produce with a substantially different production process. We compare the two production processes by estimating a model that allows for heterogeneity in technology and productivity. We treat both types of heterogeneity as intrinsically unobservable. In the model, plants choose technology before production starts. They condition subsequent input decisions on this choice. Maximum likelihood estimation is used to estimate the unconditional distribution of the technology choice, output, and inputs. The model is applied to a sample of automobile assembly plants. We control for capacity utilization, unobserved productivity differences, and price effects. The results indicate that there exist two distinct technologies. In particular, the more recent technology uses labor less intensively and it has a higher elasticity of substitution between labor and capital. Hicks-neutral productivity growth is estimated to be lower, while capital-biased (labor-saving) productivity growth is estimated significantly higher, for the new technology. Using the estimation results, we decompose industry-wide productivity growth in plant-level changes and composition effects, for both technologies separately. Plant-level productivity growth is further decomposed to reveal the importance of capital-biased productivity growth, increase in capital-labor ratio, and returns to scale.