This thesis includes one qualitative literature review, three empirical studies and one meta-analysis examining the measurement of physical activity in children. Previous research has highlighted the difficulties inherent in measuring children's habitual activity. This has lead to confusion regarding the relationships between physical activity and health in children. Recently a new type of activity monitor has been developed. Uniaxial (WAM, Computer Science Applications, Shalimar, Florida, USA) and triaxial (Tritrac, Professional Products, a division of Reining International, Madison, WI, USA) accelerometers that record temporal, frequency and intensity information of movement are now commercially available. The aims of this thesis were to evaluate the validity of these and other measures of physical activity in children, to examine the relationship between physical activity and body fat in children and to investigate the effect the mode of activity measurement has on this relationship. The main findings were: a) The Tritrac provided a significantly better estimate of scaled oxygen consumption during typical children's activities than the WAM, heart rate or pedometry; b) Physical activity, measured by the Tritrac or the pedometer, was inversely correlated with body fat and positively correlated with aerobic fitness; c) Heart rate measures of physical activity did not correlate significantly with body fat or aerobic fitness; d) Meta-analytic procedures showed a small to moderate relationship between activity levels and body fat in children; e) The strength of this relationship was heavily dependent on the method used to assess activity levels. Observation methods produced an effect size significantly higher than questionnaire or heart rate methods, though not significantly different to motion counter methods. In conclusion it appears that heart rate methods are inappropriate for the assessment of habitual activity in children. The use of motion counters or observation methods for assessing activity are recommended. Motion counter methods are particularly appropriate for medium to large sample sizes.