This paper presents a study aiming to find out the best strategy to develop a fast and accurate HMM tagger when only a limited amount of training material is available. This is a crucial factor when dealing with languages for which small annotated material is not easily available. First, we develop some experiments in English, using WSJ corpus as a test-bench to establish the differences caused by the use of large or a small train set. Then, we port the results to develop an accurate Spanish PoS tagger using a limited amount of training data. Different configurations of a HMM tagger are studied. Namely, trigram and 4-gram models are tested, as well as different smoothing techniques. The performance of each configuration depending on the size of the training corpus is tested in order to determine the most appropriate setting to develop HMM PoS taggers for languages with reduced amount of corpus available.