Tests of early reading in English speakers typically rely on the capacity to read irregularly spelled words. The approach is not feasible for more regularly spelled languages. The Reading Decision Test is a group-administered test, initially developed to study reading in Kiswahili speakers. It was subsequently validated by comparing an English language version with the more standard reading tests. The main goal of the study is to test construct validity and the secondary aim is provide evidence of invariance by sex, type of school, and grade for the Letter and Word Decision tests of the Reading Decision Test's Brazilian version. The study population comprised 487 children (259 boys), from public (226) and private (261) schools, across five grades. Confirmatory factor analysis based on a Bayesian estimator was applied to two versions of the task, both of which measured performance on both the letter-based and wordbased subtasks of the test. The models for both tasks and both versions showed good fit indices after excluding a few items. Both versions and subtests showed invariance for sex, type of school, and grade allowing simple comparisons such as t test and analysis of variance procedures to be used. The absence of a precise analysis of items in an instrument has unpredictable consequences for examiners and decision making in the clinical and educational context.