We examined how generalized and mathematics-specific language skills predicted the word-problem performance of students with mathematics difficulty. Participants were 325 third-grade students in the southwestern United States who performed at or below the 25th percentile on a word-problem measure. We assessed generalized language skills in word reading, passage comprehension, and vocabulary knowledge. In addition, we measured mathematics-specific vocabulary knowledge. To explore variation within the mathematics-difficulty population, we utilized unconditional quantile regression to determine how each of these skill sets predicted word-problem performance when controlling for computation and emergent bilingual status. Results revealed that mathematics-vocabulary knowledge significantly predicted word-problem performance at all but two quantiles (p < 0.001), with strongest predictive relations at the highest quantiles. Passage comprehension had an overall significant relation to word-problem performance (p < 0.05) that was also reflected in multiple quantiles. Neither word-reading accuracy nor generalized-vocabulary knowledge demonstrated a significant predictive relation to word-problem performance. Given the consistent relation between mathematics-vocabulary knowledge and word-problem performance across quantiles, researchers and practitioners should prioritize evidence-based mathematics-vocabulary instruction to support students' word-problem-solving skills.