In this paper, we consider the problem of modeling multiple-output nonlinear dynamic systems. We use linear model trees and ensembles of linear model trees to construct accurate models of nonlinear multiple-output dynamic systems, from input-output data. We show that it is beneficial to identify models with common structure for all outputs, utilizing interactions between the outputs. The operation of our method is demonstrated on the task of identification of the well-known two-output continuous-stirred tank reactor process. The multiple-output ensemble model has smaller complexity and comparable performance, as compared to a collection of two single-output ensemble models.