We present a method for automatically aligning a collection of similar shapes in arbitrary initial poses. By analyzing the shape collection we extract a deformation model to capture the variability in the collection. We use this information to deform an extracted template shape and use it to align pairs of shapes by direct PCA alignment. We evaluate our method on synthetically created model collections in arbitrary initial poses and demonstrate accurate results with near ground truth alignment. Our algorithm significantly outperforms existing direct PCA alignment methods, without significant computational overhead.