Effects of orthographic transparency were examined by comparing children learning to read in Albanian, Welsh, and English. Twenty Year 1 Albanian children were given a reading test consisting of a 100-word stratified sample of decreasing written frequency. They were able to read accurately 80% of the words; reading latency was a direct effect of word length (R2 =. 89); and errors tended to be mispronunciations rather than real word replacements, with hardly any null responses. These results were compared with Ellis and Hooper (2001), where the same design was used with English and Welsh children of the same age, but with 1 more year of formal reading instruction. The Albanian children read more words than the English and Welsh children, but they had longer reading latencies. Like the Welsh children, but unlike the English children, the Albanian children made more nonword errors. These results suggest that children acquire reading faster the more transparent the orthography, and that shallow orthographies promote an initial reliance on a phonological recoding strategy.