Unlike traditional methods that directly map different modalities into an isomorphic subspace for cross-media retrieval, this paper proposes a cross-media retrieval algorithm based on the consistency of collaborative representation (called CR-CMR). In order to measure the similarity between data coming from different modalities, CR-CMR first takes the advantage of dictionary learning techniques to obtain homogeneous collaborative representation for texts and images, then, it considers the semantic consistency of different modalities simultaneously and maps the collaborative representation coefficients into an isomorphic semantic subspace to conduct cross-media retrieval. Experimental results on three state-of-the-art datasets show that the algorithm is effective.