Streamfunction and velocity potential are widely used in 3DVAR systems for mesoscale data assimilation. With proper covariances or recursive filters, these 3DVAR systems provide reasonable analyses for numerical prediction models. However, when they are used to analyze smaller scale motions than a observation network can resolve, these systems may result some interesting error patterns in their incremental fields. In this presentation, theoretical discussion and numerical experiments are provided to understand these error patterns. Alternative approaches are recommended to avoid this kind of error patterns.