The focus of this study is on providing tools to enable researchers and practitioners to screen for dyslexia in adults entering university. The first aim is to validate and provide diagnostic properties for a set of seven tests including a 1-min word reading test, a 2-min pseudoword reading test, a phonemic awareness test, a spelling test, the Alouette reading fluency test, a connected-text reading fluency test, and the self-report Adult Reading History Questionnaire (ARHQ). The second, more general, aim of this study was to devise a standardized and confirmatory procedure for dyslexia screening from a subset of the initial seven tests. We used conditional inference tree analysis, a supervised machine learning approach to identify the most relevant tests, cut-off scores, and optimal order of test administration. Method A combined sample of 60 university students with dyslexia (clinical validation group) and 65 university students without dyslexia (normative group) provided data to determine the diagnostic properties of these tests including sensitivity, specificity, and cut-off scores. Results Results showed that combinations of four tests (ARHQ, text reading fluency, phonemic awareness, pseudoword reading) and their relative conditional cut-off scores optimize powerful discriminatory screening procedures for dyslexia, with an overall classification accuracy of approximately 90%. Conclusions The novel use of the conditional inference tree methodology explored in the present study offered a way of moving toward a more efficient screening battery using only a subset of the seven tests examined. Both clinical and theoretical implications of these findings are discussed.