Previous studies in continuous keystroke verification have shown that users' templates built with enrollment samples collected in multiple sessions are instrumental in reducing verification error rates. However, to our knowledge, no work has addressed how to achieve low error rates in situations where only weak keystroke templates (i.e., templates created from a single enrollment session) are available. To address the problem, we propose a framework comprising of impostor score based normalization, impostor score based rejection, and fusion. We introduce a new formulation to incorporate reject option in verification with weak templates and develop a new impostor score based rejection method called Order Statistic (OS) rejection method. We compare the performance of OS rejection method with two other impostor score based rejection methods - 1) Otsu and 2) Gaussian. We performed experiments on a large keystroke database of 1100 users. Results show that our proposed framework significantly reduces the EERs of continuous keystroke verification with weak templates and the OS rejection method achieves better error-reject trade-off than Otsu and Gaussian rejection methods. We achieved 59.97 percent to 86.74 percent reduction in average EERs compared to the individual verifiers when we used OS rejection method in conjunction with impostor score based normalization and fusion.