Automatic word spacing decides the correct boundaries between words in a sentence. Word spacing is important in Korean, and word spacing errors are frequent. Several proposed probabilistic word-spacing models resolve problems with previous statistical approaches. These models regard automatic word spacing as a classification problem similar to part-of-speech tagging. By generalizing hidden Markov models, the models can consider a broader context and estimate more accurate probabilities. The authors tested these models under a wide range of conditions to compare them with the state of the art and performed detailed error analysis of them.