Purpose: The main objective of the present study was to examine the effect of word length, word frequency and number of syllables on graphemic errors during a text-reading task via the braille code. Method: Thirty students with severe visual impairments participated in the present study. They were invited individually to read aloud a number of texts via the braille code for forty-five minutes beginning from the text that corresponded to their equivalent reading level. Results: A binary logistic regression analysis revealed a word length effect on long words but not on short or medium words indicating that despite the gradual insertion of the braille characters, the braille readers were able to efficiently process large perceptual units. Moreover, braille readers produced three times more graphemic errors in the second half of the long words compared to the first half, while the students who attended secondary education performed more graphemic errors when they read short words compared to students enrolled in primary education. Conclusion: The presence of the word length effect only on long words indicates that text reading via the braille code is mostly, but not exclusively, in parallel.