Trustworthy and reliable data delivery is a challenging task in Wireless Sensor Networks (WSNs) due to unique characteristics and constraints. To acquire secured data delivery and address the conflict between security and energy, in this paper we present an evolutionary game based secure clustering protocol with fuzzy trust evaluation and outlier detection for WSNs. Firstly, a fuzzy trust evaluation method is presented to transform the transmission evidences into trust values while effectively alleviating the trust uncertainty. And then, a K-Means based outlier detection scheme is proposed to further analyze plenty of trust values obtained via fuzzy trust evaluation or trust recommendation. It can discover the commonalities and differences among sensor nodes while improving the accuracy of outlier detection. Finally, we present an evolutionary game based secure clustering protocol to achieve a trade-off between security assurance and energy saving for sensor nodes when electing for the cluster heads. A sensor node which failed to be the cluster head can securely choose its own head by isolating the suspicious nodes. Simulation results verify that our secure clustering protocol can effectively defend the network against the attacks from internal selfish or compromised nodes. Correspondingly, the timely data transfer rate can be improved significantly.