In this paper, we report on recent experiments involving the basic POS tagging task on Greek data. Four POS taggers based on different Machine Learning approaches (Transformation-Based, Memory-Based, Hidden Markov Models and Maximum Entropy) are trained on the same corpus to perform morphosyntactic tagging. Their outputs are first examined on the basis of inter-tagger agreement and then combined to construct an ensemble in order to improve the accuracy. Three types of combination methodologies are examined. Finally, we conclude with a detailed presentation of the results along with some remarks on their limits concerning the reduction in error rate.