We present a volume modelling approach based on sequences of two-dimensional ultrasound images. Though generally applicable to arbitrary freehand ultrasound, our method is designed for the reconstruction of timevarying volumes from ultrasound images of a human heart. Since the reliability of the reconstructed data depends very much on the spatial density of ultrasound images, we apply a hierarchical modelling approach. The volume produced for each time step is represented as adaptive mesh refinement (AMR) data such that regions of low reliability in the reconstructed volume can be recognized by their coarse resolution.