Multisyllabic word reading has received little attention in existing computational models, which are designed for English. The Greek language uses mainly multisyllabic words, while its orthography is feedforward consistent and includes a stress diacritic. We present here a computational model of reading aloud Greek words and nonwords, based on the connectionist triangle model, adapted to the Greek orthography, using a novel input representation. The network displays several effects from the word reading literature and successfully assigns stress. Further investigations are underway.