The two experiments reported here were designed to examine discourse planning during language production. Following the work of Levelt (1981, 1982), participants were shown simple networks consisting of two branches, and their task was to describe the networks. The two branches of the networks differed in either length or complexity; the dependent measure was participants' decisions to describe the left or right branch first. The experiments showed that speakers preferred to describe a less complex branch before a more complex branch and preferred to describe a shorter branch before a longer branch. Levelt's minimal-load principle is invoked to explain these results, along with the principle of incrementalism (Levelt, 1989). Discussion focuses on the possibility that incrementalism in production is useful to both the speaker and the listener.