This paper aims to study the feasibility of using a smart mobile phone with an embedded accelerometer in gait pattern monitoring. The second motivation is to examine the impact of the accelerometer sampling frequency on gait analysis. A mobile phone and a standalone accelerometer sensor were simultaneously attached to subject's lower back to record walking patterns. The degree of agreement between gait features derived from two devices was assessed in terms of average error rate, normalised limits of agreement and intra-class correlation. Various agreement levels were observed for three temporal features, three root mean square features, five regularity features and two symmetry features. The downsampling data were used to examine the impact of sample intervals on the gait features. Eleven out of 13 features have normalised mean difference less than 0.1 when sample intervals were less than 50ms. To carry out a further evaluation, the features derived from the downsampling gait data were used to classify subjects with chronic pain and health subjects, and a classification accuracy of 90% was achieved. The results showed that it is feasible and reliable to assess and monitor gait patterns based on spatio-temporal gait features derived from smart mobile phones with an embedded accelerometer.