In this paper, we consider a MIMO wiretap system in wireless sensor networks (WSNs), where the confidential signal sent to the legitimate receive (Bob) may be eavesdropped by the eavesdropper (Eve). Assuming that only partial channel state information (CSI) can be obtained by the transmitter, we consider both worst case (WC) and outage-constrained (OC) robust secrecy optimizations. To solve the WC design, we propose to linearize these logarithmic determinant terms. After linearization, we tackle the CSI uncertainty using the Nemirovski lemma. Then, an alternating optimization (AO) algorithm is proposed to solve the reformulated problem. On the other hand, to solve the OC design, we transform the probabilistic constraint into safe and tractable reformulation by the Bernstein-type inequality (BTI) and large deviation inequality (LDI), and an AO algorithm is proposed. Numerical results are provided to demonstrate the performance of the proposed scheme.