This paper presents an improved time-frequency (TF) based technique for newborn EEG seizure detection. The original technique analyses successive spikes intervals of the EEG signal in the TF domain to discriminate between seizure and nonseizure activities. In this paper improvement on the original approach is achieved by using a new spike detection technique. In this technique the TF of the signal is enhanced before the actual spike detection scheme is applied. Then, two frequency slices are extracted from the higher frequency area of the TF distribution to detect the spikes. The extracted frequency slices are subjected to the smoothed nonlinear energy operator to accentuate the spike signatures. Histogram of successive spikes intervals is then used as a feature for seizure detection. In the presented technique the EEG data are segmented into 4-second epochs. A k-nearest neighbour algorithm is employed to classify the EEG epochs into seizure and nonseizure groups. The performance of the presented technique is evaluated using the EEG data of five neonates.