Procedural flexibility is an important skill for algebra. Although prior work has focused on measuring students' procedural flexibility using arithmetic problems, word problems may also capture students' flexibility because of their open-ended nature. To date, no published study has examined the use of word problems as another measure of procedural flexibility. The present study aims to establish flexibility on fraction word problems as a predictor of algebra learning and to determine whether flexibility demonstrated on fraction arithmetic versus fraction word problems differentially predicts two types of algebra learning (e.g., algebraic feature knowledge and algebra equation-solving). Middle-school students (N = 350) completed fraction arithmetic problems, fraction word problems, and algebra measures at the start and end of the school year. We coded fraction arithmetic and word problems for demonstrated procedural flexibility. Path models showed that overall flexibility significantly predicted both algebraic feature knowledge and equation-solving. When examining flexibility by type, procedural flexibility on arithmetic problems significantly predicted end-of-year algebraic feature knowledge and algebra equation-solving, whereas procedural flexibility on word problems significantly predicted algebraic feature knowledge. Procedural flexibility on arithmetic problems was still predictive of performance when accounting for fraction magnitude skill. These findings build on prior literature that establishes procedural flexibility as a crucial predictor of algebra learning and extends measures of procedural flexibility to include word problems.