Automated, image based high-content screening has become a fundamental tool for scientists to make discovery in biological science. Modern robotic fluorescence microscopes are able to capture thousands of images from massively parallel experiments such as RNA interference (RNAi). As such, efficient methods are required for automatic cellular phenotype identification capable of dealing with large image data sets. In this paper we applied the Curvelet transform for image feature description and Random Subspace ensemble (RSE) for classification. The Curvelet transform as a new multiscale directional transform allows an almost optimal nonadaptive sparse representation of objects rich with edges. The RSE contains a set of base classifiers trained using randomly drawn subsets of curvelet features. The component classifiers are then aggregated by the Majority Voting Rule. Experimental results on the phenotype recognition from three benchmark fluorescene microscopy image sets (RNAi, CHO and 2D Hela) show the effectiveness of the proposed approach. The ensemble model produces better performance compared to any of individual neural networks. It offers the classification rate 86.5% on the RNAi dataset, which compares favorably with the published result 82%, and the results on the other two sets of fluorescence microscopy images also confirm the effectiveness of the proposed approach.