Automated analysis and grading of the neonatal EEG has a potential to assist clinical decision making for neonates with hypoxic-ischemic encephalopathy. This paper proposes a method to grade the degree of abnormality in hour-long segments of neonatal EEG. The HMM-based speaker diarization approach is employed to segment and cluster the neonatal EEG into homogeneous states. Several features are proposed to characterize the resultant state sequence to provide a single measure for a complete hour-long EEG recording. These features aim at capturing both the statistics of the state durations (e.g. average state duration, average number of segments), and any patterns contained in the sequentiality of the obtained states (e.g. permutation entropy, entropy rate). Statistical analysis indicates that the proposed features contain discriminative information for the task of automated neonatal EEG grading. Unlike other studies, the developed framework of the EEG 'diarization' provides an easy and intuitive interpretation of the computed features, which is a clinically important aspect. © 2014 IEEE.