In earlier experiments, the establishment of baseline relations using a delayed matching to sample (DMTS) trial format resulted in a high likelihood of forming equivalence classes (yield) that had many-to-one or one-to-many training structures. Because these training structures are associated with high yields, training structures instead of the delay could have contributed to the results. Equivalence classes are less likely to form when they are larger and have linear series training structures. Thus, we studied the effects of delay on the likelihood of forming larger five-member equivalence classes with three-node linear series training structures. In different groups, training was conducted using 0.1-s, 3-s, or 12-s DMTS trial formats. Intermediate yields were obtained when training was conducted with 100- and 3-s delays, while all participants formed classes when training was conducted with a 12-s delay. These results support and extend earlier findings on DMTS in stimulus equivalence research. The results also indicate that an experience with 12-s DMTS in AB training was more favorable in acquisition of BC conditional discriminations than was 0.1- and 3-s DMTS, but no clear difference between different delay durations was seen on the number of trials necessary to establish the totality of baseline conditional discriminations.