This paper considers the problem of continuous-time model identification with arbitrary time-delay from irregularly sampled data. The proposed method estimates the plant and the time-delay in a separable way, when estimating one of them, the other is assumed to be fixed. More precisely, the plant is estimated by the iterative instrumental variable SRIVC method while the time-delay is estimated by the Gauss-Newton method. Because of the nonlinear relationship between the loss function and the time-delay, a low-pass filter is employed to extend the global convergence region for the time-delay estimation. Numerical examples are presented to illustrate the properties of the proposed method.