Using eight-letter words, masked form-priming is examined with more extreme versions of transposed primes than are normally considered. Strong priming effects are observed when as few as two out of eight letters are correctly positioned within the prime, indicating that the word recognition system is surprisingly flexible with regard to letter position. Three prominent models of letter position coding are considered in light of the observed data. The comparison between models focuses on the assumptions and mechanisms within each. Strengths and weaknesses are identified for all approaches. Some evidence is found in support of differential weighting of letter positions, although the more specific question of whether it is exterior or initial letters that are most crucial to word recognition remains unresolved.