Impairments in object naming have contributed to a view held in neurology that many higher brain functions can be understood as the result of sequential stages of processing connected by limited-channel pathways. However, this standard model cannot readily account for some specific types of impairments, nor for important variables such as speed. The author reinterpreted the classic functional/neuroanatomic stages as examples of distributed, parallel processing in neural networks with massively parallel interconnections. Simulated 'lesions' between the stages of this model could reproduce the impairment of real patients with such (presumed) lesions. Such models have to be strongly considered in understanding higher cerebral function and its disorders; empirical testing and future directions are discussed.