The K-complex detection task is approached by first extracting morphological features that quantify the visual recognition criteria used for both acceptance and rejection of candidate waveforms. The features are based on amplitude and duration measurements. These features are used as the inputs of multivariate discrimination methods. The performance of Fisher's linear discriminant with multilayer feedforward neural networks (MLFNs) in discriminating the K-complex and background EEG is compared. The results show that the use of the MLFN on feature information can provide a reliable K-complex detection with significantly better performance than that of the linear discriminant. This difference in performance can be seen on the receiver operating characteristics curves that show the true positive against the false positives.