We consider cases where reality is best described by a continuous model, and where data are sampled at discrete points in time. Then an exact transformation of the continuous model into a discrete one, or vice versa, is typically very complicated. Simplified transformations might produce great errors if the sampling interval for the time series is approaching natural periods or time constants of the system being modelled. For such problematic cases we discuss implications for a system dynamics, traditional discrete model econometrics, and Bayesian statistical methods.