This study investigates the relative roles of internal and external letter features in word recognition. In Experiment 1 the efficacy of outer word fragments (words with all their horizontal internal features removed) was compared with inner word fragments (words with their outer features removed) as primes in a forward masking paradigm. These forward masked primes were followed by a word to be read aloud. Outer word primes presented for longer durations produced significantly faster naming responses than inner primes. Outer parts of words appear to provide more relevant information for lexical access at an earlier stage than inner fragments. In Experiment 2 words with only external features were named correctly on 96% of occasions compared with 52% of words with only their inner features presented. This indicates much greater information content in the periphery of a word (despite having a reduced area of print available: 45% compared to 55%). Multiple regression analyses controlling for "guessability" (from data in Experiment 2) still produced significantly faster reaction times in the outer relative to the inner priming condition for longer prime durations. These experiments demonstrate that first, the most informative letter features are concentrated in the peripheral region of words; and second, even controlling for this effect, readers appear to have a bias towards analysing outer features of a word before inner features.