The use of peer assessment to evaluate students' writing is one recommended method that makes writing assignments possible in large content classes (i.e., more than 75 students). However, many instructors and students worry about whether students of all ability levels are capable of helping their peers. We examine how ability pairing (e.g., high-ability student with high-ability student versus high-ability student with low-ability student) changes key characteristics of feedback to determine which pairings are likely to benefit students most. A web-based reciprocal peer-review system was used to facilitate the peer review of students' writing of two papers. Over 1,100 comments given to writers from their peers were coded for several relevant categories: type of feedback, type of criticism, focus of problem, focus of solution, and implementation. Overall, creating peer-review groups such that students receive feedback from someone of a dissimilar ability appeared to be most beneficial. High-ability writers received similar kinds of feedback from high-ability versus low-ability peers. By contrast, the low-ability writers received more comments that identified problems focusing on substance issues from high-ability reviewers. In addition, the low-ability writers implemented a higher percentage of the comments from the high-ability reviewers.