Background <p>Among neurotypical children working memory and language development are closely related, but the longitudinal relationship between these factors is unknown in autism. This study had the goal of evaluating potential bidirectional relationships between these domains.</p> Methods <p>A longitudinal design included language and visuospatial working memory assessments across three time points with 160 autistic preschoolers. Time points were one year apart. A cross-lagged panel model (CLPM) using Full-Information Maximum Likelihood (FIML) was used to evaluate longitudinal associations between performance on a working memory task (i.e., the Hide and Seek Task) and language scores (i.e., verbal development quotient).</p> Results <p>Cross-lagged associations were significant for paths from working memory at the first time point to language at the second time point (β = .31, <i>p</i> = .01) and from language at first time point to working memory at the second time point (β = .40, <i>p</i> &lt; .001). Cross-lagged paths from the second to third time point were not statistically significant in either direction.</p> Conclusions <p>This study provides evidence of a bidirectional relationship between working memory and language in autistic toddlers and preschoolers among younger autistic preschoolers. These results highlight working memory as a potential early assessment and intervention domain for improving language outcomes in autism as well as the role of language delays as a factor related to reduced working memory skills later in development.</p>