Cognitive assessment in natural group settings facilitates data collection but poses threats to the validity. In this study, tablet-based working memory (WM) tasks, the counting span, and reading span were used in predicting 12-year-old children's (N = 837) scholastic skills and fluid intelligence in a classroom with environmental noise. WM tasks had excellent internal consistency, correlated with scholastic skills, and accounted for more of the variance in cognitive performance (grade point average, fluid intelligence, scholastic skills) compared with individually administered (n = 190) digit span task. Furthermore, the multilevel analysis revealed that compared with the classrooms with no noise, when naturally occurring speech or nonspeech types of environmental noises were present during assessment, WM scores or the reliability estimates were not lower. In contrast, when both types of noises were present, the relationships between some of the WM and achievement scores were even stronger. Thus, assessments in natural classroom contexts may promote revealing the individual differences in WM.