The purpose of this study was to investigate literacy development beyond the early school years. We investigated mean-level and rank-order changes in two reading (word decoding speed and reading comprehension) and two spelling (word and pseudoword spelling accuracy) tasks during a 4-year period from beginning to the later grades of elementary school, and examined whether rank-order changes in literacy skills could be predicted from verbal working memory, phonological awareness, and morpho-syntactic awareness. The sample consisted of Croatian speaking children, poor (n = 50) and good (n = 61) readers. Croatian is a language with a highly transparent orthography but a rather complex grammar. Word decoding speed and word spelling accuracy improved significantly over the study period, word decoding speed more so among good than among poor readers, supporting the Matthew effect hypothesis. Literacy skills were moderately stable, with the highest stability coefficients obtained for word decoding speed, and the lowest for pseudoword spelling. The predictors of rank-order changes varied across literacy outcomes, but did not differ for poor and good readers. Morpho-syntactic awareness predicted rank-order development of all outcomes except for word spelling. Phonological awareness predicted rank-order development of reading comprehension and word spelling. Verbal working memory predicted rank-order development of word spelling only, and at a low level. The finding that the more language related cognitive variables, i.e., morpho-syntactic and phonological skills, were stronger predictors of literacy development than working memory, for both poor and good readers, suggests remedial focus on these more predictive variables.