Typically, a Korean Part-of-speech (POS) tagger takes the inputs that are produced by a separate Korean word spacer. However this pipeline approach has an obvious flaw of error propagation, since the POS tagger cannot correct word spacing errors. In this paper, we describe a joint model for Korean word spacing and POS tagging using structural SVM to avoid error propagation and improve word spacing by utilizing POS information. In the case of a pipeline approach, we could achieve a 96.77% morpheme-based F-measure for POS tagging. Using the joint model, we could achieve a 96.99% morpheme-based F-measure for POS tagging. Experimental results show that the joint model outperforms the pipeline approach.