We introduce a model that maps variable-length word utterances to a word vector space using convolutional neural networks. Convolutional networks are a rich class of architecture that, through many nonlinear layers, can model complex functions of their input. Our approach models entire word acoustics rather than short windows as in previous work. We introduce the notion of mapping these word inputs to a word vector space, rather than trying to solve the mas- sively multi-class problem of word classification. Regressing to word vectors offers many opportunities for further work in this domain, as many techniques exist to learn word vectors for different notions of word similarity. We experiment on hundreds of hours of broadcast news, and demonstrate our model can accurately recognize spoken words. Further, we use our model to build features for the SCARF speech recognition system and achieve an improvement in large vocabulary continuous speech recognition over a baseline system.