Connectionist networks have been used extensively in the domain of visual word recognition and reading. This paper presents a connectionist model that builds on the split-fovea model (Shillcock and Monaghan, 2003). The split-fovea model integrates into its architecture the facts that the brain and the fovea are both split. To simulate the split fovea, the model has 2 input layers. This allows for words to be presented to the network at different fixation positions. The paper compares two networks: 1. A control net for which words are presented with the same frequency at all fixation posi- tions during training. 2. A fixation net for which the frequencies for words at different fixation positions are determined by actual data of people while reading. The aim of the paper is to establish any differences occuring in the fixation net due to the different fixation positions.