This paper presents an objective comparison between two recent constitutive models employing the concept of the hardening memory surface to predict the high cyclic loading behaviour of granular soils. The hardening memory surface is applied to the well-known Severn-Trent sand and the SANINSAND04 constitutive models. While the addition of the new model surface (the memory surface) leads to enhanced model capabilities, slight differences in the implementation can lead to different model performances and simulations. This paper describes the differences between the two implementations and highlights the most relevant modelling ingredi-ents to predict particular features of the cyclic soil behaviour. This paper will help the reader in selecting the most suitable model and related ingredients for a particular geotechnical application.