Medical knowledge systems are computer systems which contain medical knowledge and use it to reason about patients. Although the first was described over three decades ago, few have passed into routine use, compared to the large numbers of laboratory databases, CT scanners or computerised electrocardiogram (ECG) interpreters, which now process over 50 million ECGs per annum. This paper considers the lessons that we can learn from computerised ECG interpreters, and discusses six major issues: establishing closer dialogue between system developers and users, evaluating the impact of systems on users and their problems, integration with patient record systems, assembly of public, validated medical knowledge bases, reducing the emphasis on novel reasoning methods and considering the legal, regulatory and educational implications of this technology. Specific research topics are proposed under each of these headings.