This report is a follow-up to a previous paper which described seizure rate changes with central cortical EEG feedback training in 8 poorly controlled epileptic subjects. Data examined here include associated training compliance and performance, sleep EEG spectra, clinical EEG and anticonvulsant blood levels. The study employed a double-cross-over, single blind ABA design applied to two subgroups of epileptic patients. Both groups had in common two training periods (A1, A2) in which either 12--15 c/sec (subgroup I, n = 4) or 18--23 c/sec (subgroup II, n = 4) was reinforced in the absence of 6--9 c/sec, movement or epileptiform discharge, and one training period (B) in which 6--9 c/sec was reinforced in the absence of 12--15 or 18--23 c/sec as well as movement and epileptiform discharge. Training periods occurred primarily in the home and lasted 3 months. Compliance with training instructions and response acquisition were demonstrated. Overall anticonvulsant blood levels were low and unrelated to EEG or seizure changes. Clinical EEG findings corresponded to sleep EEG and seizure rate outcomes. Power spectral analysis of sampled non-REM sleep from all-night EEG recordings obtained after each training phase indicated contingency specific changes which were limited to sensorimotor recordings in subgroup I and corresponded to the pattern of seizure rate changes in this group. EEG changes were also limited to sensorimotor cortex in subgroup II, but were linear and paralleled a progressive decrease in seizure rate. Both groups, however, showed the same pattern of EEG changes with seizure reductions; low and high frequencies were reduced and intermediate, rhythmic frequencies increased. Correlational analysis confirmed this relationship. The pattern, duration and topographic specificity of these changes suggested a normalization of sensorimotor EEG substrates related to the EEG feedback traning.