This paper proposes a compressed sensing method based on overcomplete learned dictionaries for experimental 3D ultrasound (US) imaging. Two undersampling patterns suited for 3D US imaging are investigated in ex vivo 3D US volume acquisitions: a spatially uniform random acquisition and a line-wise random acquisition. The overcomplete dictionary was learned using the K-SVD algorithm on patches extracted from 3D US datasets. The CS reconstruction problem was solved through the l1 minimization using the spectral projected gradient algorithm and was performed by removing 20% to 80% of the original samples according to the two undersampling patterns. Besides the K-SVD dictionary we include the Fourier basis and the discrete cosine transform to our experiments for comparison. The approach is evaluated on 3D US experimental data acquired from ex vivo pig brains and sheep hearts. Reconstructions from 50% of the samples of the original 3D volume show little information loss in terms of normalized root mean squared errors.