High-frequency retrofiltration is a promising technique of flux enhancement, which should contribute to the development of cross-flow microfiltration. Most published models have assumed cake deposit in the forward filtration phase, incomplete cleaning by backwash, and delays in cake formation and erosion. However, they were fitted to the overall net flux since the transient signals could not be monitored. The experimental set-up of this work allowed obtaining the forward flux response to periodical backpulsing and measuring the average reverse flux. The tested suspensions were bentonite in tap water and biologically treated wastewater. Average reverse flux was accurately measured by two different methods leading to the same results. Moreover most parameters, introduced in the published models, were experimentally measured and not fitted to the results as previously, e.g. the cleaning efficiency, which has been an adjusted parameter, was measured by two methods leading to the same value. The forward flux responses are not in agreement with the models. Moreover, the models do not accurately predict the average reverse flux when the suspension interacted with the membrane. The average reverse flux could even be higher than the forward flux, which could hamper the economical achievement of the technique. However, the optimal frequency was fairly well predicted by all the models for both the suspensions. A method of determination of the net flux and the optimal conditions was proposed. This method is based on the forward flux response to one pulse obtained when the system is well stabilised; it accurately predicted the net flux.