The rapidly changing nature of the EEG signal during an epileptic seizure presents significant problems for both linear and non-linear methods of signal analysis that require stationarity of the signal. This paper reports application of time-frequency analysis to human temporal lobe seizure activity recorded using subdural grid and depth electrodes. The matching pursuit method allows time-frequency decompositions of EEG signals during seizures. The energy density plots computed for whole seizures show changing frequency composition of the signal during a seizure. The same analysis can be applied to signals generated by the Duffing equation exhibiting either limit cycle or chaotic behavior to compare relative complexity of signals. A quantitative measure of complexity of signal based on the time-frequency decomposition is introduced. This measure is defined as the number of time-frequency atoms needed to account for 90% of the energy of the signal. There is relatively high complexity in the initial period after seizure onset, followed by lower complexity when rhythmic activity predominates, and later increased complexity again during the intermittent bursting period before seizure termination.