Several recent studies have quantified abnormalities in Parkinsonian gait. However, few studies have attempted to quantify the regularity of body motion during walking in patients with Parkinson's disease. The aim of the paper was to characterise body motion patterns in healthy, elderly subjects and patients with Parkinson's disease during walking. Body motion was recorded during walking for 16 patients with Parkinson's disease and ten healthy, elderly subjects using a tri-axial accelerometer device. To characterise the body motion patterns, time-frequency patterns of the body acceleration signal were estimated using a matching pursuit algorithm. Data from the study showed that the healthy, elderly subjects and patients with Parkinson's disease had different time-frequency patterns. The time-frequency patterns were classified into four distinct patterns based on their time durations: vertical (< 0.15 s), circular (0.15-0.5 s), short horizontal (0.5-2.0 s) and long horizontal (> 2.0 s). The data showed that the energy of the long horizontal patterns, representing long-term smooth and regular (rhythmic) activities, significantly decreased, but the energy of the circular patterns, representing irregular activities, increased in the patients with mild Parkinson's disease, compared with those of the healthy, elderly subjects (p < 0.01). Futhermore, these features were seen more clearly in the body motions of severe case patients than is that of mild case patients. It was concluded that these differences are probably due to a lack of ability to control normal and smooth movement is Parkinson's disease.