Due to practical limitations such as inter- and intraleaf transmission, nondivergent leaf end design, and leaf scatter, multileaf collimators (MLCs) are unable to accurately produce the ideal fluence patterns generated by inverse planning systems. Consequently, low dose regions receive substantially more radiation than they would with an ideal MLC that could generate the desired fluence pattern. Previous work by others has found that the discrepancy between desired and actual fluence patterns produced by an MLC increases rapidly with increasing complexity of the desired fluence map. In addition to the complexity of individual fluence maps, other parameters can contribute to the overall complexity of a treatment plan, most notably the number of beams. In this work, we investigate the effect of beam number on critical structure sparing for dynamic MLC delivered intensity modulated radiation therapy. Six cases from each of two challenging clinical sites, previously irradiated head and neck and paraspinal metastasis, were planned with the goal of minimizing the spinal cord dose. Plans were developed for five to 27 beams. All plans were renormalized such that the target volume receiving the prescription dose was the same for all plans of each site. For each case, we calculated the spinal cord D0.5 cm3 (the dose such that 0.5 cm3 of normal tissue receives greater than or equal to D0.5 cm3), normal tissue D1 cm3, the normal tissue mean dose, and the standard deviation of dose in the planning target volume (PTV). For the head and neck cases, the mean increase in spinal cord D0.5 cm3 between seven and 27 beam plans was 10% of the prescription dose, whereas for the paraspinal case, the increase was 2.6%. For the head and neck cases, the mean decrease in normal tissue D1 cm3 between seven and 11 beam plans was 2.6% and was constant for more than 11 beams. For the paraspinal cases, the mean decrease in normal tissue D1 cm3 between seven and 27 beam plans was 3.7%. The mean normal tissue dose was approximately independent of the number of beams for both sites. For the head and neck cases, the PTV standard deviation was independent of the number of beams, while for the paraspinal cases it decreased by an average of 1.8% from seven to 27 beams. Calculations for seven and 27 beams in which the MLC transmission was varied from 0% to 2% demonstrated that the increase in spinal cord D0.5 cm3 with increasing number of beams is largely due to MLC transmission, which is not included in the optimization. Increasing the number of beams increased the critical structure dose, although decreasing beam number results in increasing normal tissue D1 cm3 and target dose heterogeneity. The optimal tradeoff is dependent on the clinical situation, but seems to be seven to nine beams. Beam geometry optimization may reduce the number of beams required to provide adequate target coverage, thus limiting critical structure dose. © 2007 American Association of Physicists in Medicine.