We set out to address the adequacy of dyslexia screening in Dutch and non-western immigrant children, using the Dutch Dyslexia Screening Test (DST-NL) and outcomes of the Dutch dyslexia protocol, both of which are susceptible to cultural bias. Using the protocol as standard, we conducted an ROC (Receiver Operating Characteristics) analysis in Dutch and immigrant third, fifth, and seventh graders, combining a cross-sectional and longitudinal design. Sensitivity and specificity increased with grade, but were non-significant for various subtests in the lowest grade, suggesting considerable non-convergence between the two measures. Effective subtests in all grades, presumably not strongly influenced by Cultural Background or Word Lexicon, were One-Minute Reading, Non-Word Reading, and Nonsense Passage Reading. In a multilevel analysis, cultural background, dyslexia diagnosis, parental education, and grade of first assessment were predictors of subtest performance. In a second analysis, Word Lexicon was added as a proxy of knowledge of the Dutch language and culture. After controlling for Word Lexicon, cultural background became significant for most subtests, suggesting the presence of cultural bias. Subtests assessing technical literacy, such as One-Minute-Reading, Non-Word-Reading, One-Minute-Writing, or Two-Minutes-Spelling, showed more convergence between the two assessments. Less-effective subtests were Naming Pictures, Backward Digit Span, and Verbal and Semantic Fluency. It is concluded that the DST-NL and the standard protocol do not show complete convergence, notably in the lower grades in the multilingual pupil group of our cohort, mainly because dyslexia and literacy difficulties are hard to disentangle.