The purpose of this study was to compare the semi-prone (SP) and sitting (SIT) training positions with respect to head control in children with cerebral palsy, before and after 5 weeks biofeedback training using a head position trainer (HPT). Four children were randomly assigned to each of two training groups: (a) SP on a prone board inclined 45 degrees above the horizontal and (b) SIT in their personal wheelchair and orthotic device. The HPT, secured to the child's head, controlled a video-cassette player, turning it off when the head deviated beyond 25 degrees from the vertical (termed an error). The time in error and the number of errors during test periods of 3 minutes, without feedback and completed in both the SP and the SIT positions, were determined immediately before and after, and at 16 weeks after training. The SIT trained group performed significantly better immediately post-training in three of four comparisons (P < 0.01), but the groups performed similarly in the other eight comparisons--four immediately pre-training and four at 16 weeks post-training (P > 0.05). Post-training scores for the total group (n = 8) were significantly improved over pre-training scores, regardless of the test position or the criterion measurement (P < 0.05). Biofeedback training with a HPT can be effective in either the SIT or the SP positions, with improvement lasting at least 16 weeks after training is discontinued.