Multiple-choice questions for which more than one of the alternatives given could be correct were answered either by responding to each alternative (putting a tick or cross as appropriate) or by responding to subsets of alternatives (using code letters to denote the appropriate subset). In Experiment 1 fewer errors were made when code letters were used, but the range of subsets available provided information about correct answers. Experiment 2 eliminated this information by increasing the number of codes and extended the investigation to two levels of question difficulty answered by subjects from three levels of verbal ability. Response mode did not affect the accuracy of answering easy questions, but with harder questions coded answers were both slower and more error-prone than responses made to each alternative. These performance decrements were greatest for subjects in the lowest ability grouping. It is suggested that computer software can be adapted to meet the requirements of the human information processing system more easily than people can adapt to fit the requirements of computers.