Abstract. Power electronic three-level AC-DC converters connecting a utility grid with DC supplies are widely used in renewable energy sources. For such distributed systems, the quality of energy is very important. Hence, control algorithms of an AC-DC converter have to provide not only a sinusoidal shape of grid current, but also stable operation during transient states in the grid, e.g. voltage dips or voltage unbalance. This paper deals with the problem of robustness of AC-DC control methods under disturbed grid voltage. Three modern predictive control algorithms are presented and investigated under grid voltage distortions. Experimental results from a laboratory test setup illustrating properties of the described methods are shown. Conventional linear control methods, such as VOC and DPC-SVM, are designed to control only the fundamental component of current. In order to fulfill GCRs and achieve robustness under grid voltage distortion, these linear methods need additional compensation algorithms that, in turn, increase their complexity [16]. There are many publications concerning improvements of the classical control methods in order to ensure proper converter operation under grid voltage distortions [17][18][19]. However, there are only few that relate to non-linear or predictive control methods of AC-DC converters under grid voltage distortion [20,21]. This paper presents three modern non-linear control algorithms. The first one, direct power control three levels three areas (DPC-3L-3A) [13], is an extended version of the look-up table DPC with new optimal error plane distribution as well as with transfer of the error vector into a new coordinate system. Selection of the converter voltage vector is based on the error vector angle, which makes it possible to eliminate the inconvenience of using comparators. The second method presented here, called predictive control of current of three-level converter (PCi-3L) [22], belongs to the group of finite control set model predictive control methods (FCS-MPC). In the PCi-3L strategy the switches are controlled directly without modulation. This is performed by online minimizing of specific cost function J. Depending on the definition cost function, we can control many different variables, e.g. current, dc-link voltage, switching frequency, etc. Cost function is calculated for all converter voltage vectors. Minimum value of J determines the optimal switching combinations.