This paper investigates intelligent reflecting surface (IRS) enabled non-line-of-sight (NLoS) wireless sensing, in which an IRS is dedicatedly deployed to assist an access point (AP) to sense a target at its NLoS region. It is assumed that the AP is equipped with multiple antennas and the IRS is equipped with a uniform linear array. The AP aims to estimate the target's direction-of-arrival (DoA) with respect to the IRS based on the echo signal from the AP-IRS-target-IRS-AP link. Under this setup, we jointly design the transmit beamforming at the AP and the reflective beamforming at the IRS to minimize the DoA estimation error in terms of Cram\'er-Rao lower bound (CRLB). Towards this end, we first obtain the closed-form expression of CRLB for DoA estimation. Next, we optimize the joint beamforming design to minimize the obtained CRLB, via alternating optimization, semi-definite relaxation, and successive convex approximation. Finally, numerical results show that the proposed design based on CRLB minimization achieves improved sensing performance in terms of lower estimation mean squared error (MSE), as compared to the traditional schemes with signal-to-noise ratio maximization and separate beamforming designs.