A script that was systematically faded from end to beginning was used to teach peer initiations about recently completed, current, and future activities. The effectiveness of the script‐fading procedure was assessed via a multiple baseline design across 4 children with autism. During baseline, the children seldom initiated to peers, although all had previously acquired some functional expressive language and sometimes spontaneously addressed adults. When the script was introduced, peer initiations increased, and as the script was faded, unscripted initiations increased. With the minimal written prompts available in the final fading steps, initiations generalized to a different setting, time, teacher, and activity; and for 3 of the 4 children, peer initiations were maintained at a 2‐month follow‐up. After the script was faded, the participants' levels of peer initiations were within the same range as a normative sample of 3 nondisabled youngsters. The script‐fading procedure enabled children with severe social and verbal deficits to practice context‐specific, peer‐directed generative language that was not prompted by adults or peer c