This paper describes an experiment in which subjects developed a spreadsheet model working alone, in teams of two (dyads) or in teams of four (tetrads). The goal was to determine whether synchronous development could reduce errors overall and could reduce different types of errors. Synchronous group development reduced errors compared to individual development, but only moderately. Group development was best for omission errors and Eureka logic errors. It was not good for reducing Cassandra logic errors. Mechanical errors tended to happen too quickly for team members to recognize.