In order to reduce the costs of industrial testing of analog and Radio Frequency (RF) integrated circuits, a widely studied solution is indirect testing. Indeed, indirect testing is based on learning-machine algorithms to train a regression model that links the space of low-cost indirect measurements to the space of performance parameters guaranteed by datasheets, thus relaxing the constraints on expensive test equipment. This article explores the potential benefit of using ensemble learning in this context. Unlike traditional learning models that use a single model to estimate targeted parameters, ensemble-learning models involve training several individual regression models and combining their outputs to improve the predictive power of the ensemble model. Different ensemble methods based on bagging, boosting or stacking are investigated and compared to classical individual models. Experiments are performed on three RF performances of a LNA for which we have production test data and model quality is discussed in terms of goodness-of-fit, accuracy and reliability. The influence of the training set size is also explored. Finally, the efficiency of classical and ensemble models is compared in the context of a two-tier test flow that permits to tradeoff test cost and test quality.