<p>In this paper, a new operation mode switching method is proposed for a blind decision feedback equalizer (DFE) aiming to avoid the problems of using fixed performance threshold levels and thus provide a faster convergence rate. The operation mode switching based on a priori selected (fixed) performance thresholds is commonly done by trying different threshold levels which is not practical and implies a lack of efficiency. The use of fixed thresholds degrades the performance of blind equalizers because system parameters are unknown, and it can even lead to their convergence failure. In contrast, the new method, called the adaptive threshold method, switches the DFE operation modes by comparing the convergence state of the complete equalizer including both feedforward (FFF) and feedback (FBF) equalizers, with the convergence state of the FBF equalizer alone. In this scenario the convergence state of the complete equalizer is quantified by its output mean square error (MSE), while the convergence state of the FBF is quantified by a <i>convex functional</i> defined as the weighted sum of the kurtosis number of the given signal constellation and the a posteriori error of the FBF, with the weighting factor being a quantity inversely proportional to the <i>correntropy coefficient</i> of the FBF. Based on these two performance measures, the DFE switches operation modes by making the comparison between the output MSE and the convex functional which acts as an adaptive performance threshold level in respect of the MSE. The efficiency of the adaptive threshold method is evaluated by simulations using different QAM signals and equalization scenarios. The results of comparison tests between the adaptive threshold method and the one using fixed thresholds have shown the superiority of the proposed method.</p>