We present an analysis of reading errors on a manually transcribed dataset of Grade 3 students (N=595) who read aloud level-appropriate passages in English and Hindi. Substitutions are categorised as word or non-word errors. Further, substitutions are analysed using grapheme/phoneme sequence matching and assigned typical reading error types such as initial part matches, final part matches and scaffolding errors. We compare the distribution of error categories for the two languages and discuss underlying language-dependent decoding strategies. We also apply the analysis methodology to identify the percentage scaffolding errors per utterance given its usefulness for reading assessment. Finally, we extend our work to its practical application for diagnostics by testing an automatic phone recognition system on our task.