This paper describes a framework for the analysis of accelerometer data as part of research undertaken in preparation for a clinical trial involving ambulatory monitoring of elderly rehabilitation patients. In particular we examine the response of side-mounted accelerometers to various gait patterns and attempt to establish a relationship with a biomechanical model for human gait. We explore the use of a linear predictive (LP) model as a basis for identifying key harmonic frequencies in the accelerometer response signals and use these harmonics to relate measured data back to harmonic predictions from the biomechanical model.