Purpose: This study compared the effectiveness of two grammar treatment procedures for children with specific language impairment. Method: A double-blind superiority trial with cluster randomization was used to compare a cueing procedure, designed to elicit a correct production following an initial error, to a recasting procedure, which required no further production. Thirty-one 5-year-old children with specific language impairment participated in 8 small group, classroom-based treatment sessions. Fourteen children received the cueing approach and 17 received the recasting approach. Results: The cueing group made significantly more progress over the 8-week treatment period than the recasting group. There was a medium-large treatment effect in the cueing group and a negligible effect size in the recasting group. The groups did not differ in maintenance of treatment effects 8 weeks after treatment. In single-subject analyses, 50% of children in the cueing group and 12% in the recasting group showed a significant treatment effect. Half of these children maintained the treatment effect 8 weeks later. Conclusion: Treatment that used a structured cueing hierarchy designed to elicit a correct production following a child's error resulted in significantly greater improvement in expressive grammar than treatment that provided a recast following an error.