Data mining is performed to support decision making, but many of the most powerful techniques such as neural networks, or ensembles produce opaque models which are not comprehensible for a human. The lack of interpretability is an obvious disadvantage since decision makers require some sort of explanation before taking action. To achieve comprehensibility, accuracy is often sacrificed by the use of simpler models such as decision trees. Another alternative is, however, to extract rules from the opaque model. We have previously suggested a rule extration algorithm namned G-REX. In this study we further evaluate G-REX on estimation tasks. Two new representation languages are compared to the original, using eight publicly available datasets. The extracted rules are compared to two standard techniques producing comprehensible models; multiple linear regression and the decision tree algorith C&RT. The results show that G-REX outperforms the standard techniques when an appropriate representation is used.