Our accuracy results are competitive with a median human cardiologist; and also competitive to a learner trained on the same dataset, but with the application of copious quantities of background knowledge. Given that we did not use any background knowledge, the results are good. As for comprehensibility, our system might be an appropriate ``first cut'' classifier that with just 24 rules obtains 60 per cent accuracy. Further investigations of individual classifiers shows some correspondence to rules found in commercial ECG classifiers and also show some correspondence to findings by de Chazal.