A model for predicting medication name confusion is described. Many medication errors are caused by look-alike and sound-alike medication names, yet few procedures exist to ensure the safety of new drug nomenclature or to identify confusingly similar names from within existing databases. In this study, three automated, quantitative measures of orthographic similarity (i.e., similarity in spelling) were identified (bigram similarity, trigram similarity, and Levenshtein distance). The relationship between orthographic similarity and the likelihood of a medication error was examined. For each measure of similarity, the frequency distribution of similarity scores for pairs of drug names previously reported to cause confusion (error pairs) was compared with the distribution of similarity scores for control pairs randomly selected from the general index of USP DI—Volume I: Drug Information for the Health Care Professional. Then, three parallel, unmatched case-control studies were conducted to discover whether similarity was a significant risk factor for medication errors. Finally, on the basis of the three similarity measures, tests for predicting confusion were developed and evaluated. For each similarity measure, the frequency distribution of error pairs was significantly different from that for control pairs, and orthographic similarity was a significant risk factor for medication errors. Pairs of names whose measures of similarity exceeded preset thresholds were between 25 and 523 times more likely to be involved in a medication error than pairs whose similarity did not exceed these thresholds. A prognostic test that correctly identified 91% of all pairs as either errors or controls was developed. This test had a sensitivity of 84% and a specificity of 99%. Automated measures of similarities between medication names can form the basis of highly accurate, sensitive, and specific tests of the potential for errors with look-alike and sound-alike medication names.