This letter considers the channel estimation problem when a reconfigurable intelligent surface (RIS) is used to aid wireless communications between a base station (BS) and an user equipment (UE). Unlike prior works, we assume that the directional or isotropic interference from other UEs is present at the location where the RIS is located. To address the problem, we make use of a two-stage training phase that, for any RIS configuration, operates as follows. In the first phase, the UE is not transmitting and the BS estimates the interference contribution that reaches it. This is then subtracted in the second phase during which the cascaded channel from the UE to the BS through each RIS element is estimated. In doing so, no a priori knowledge of the spatial correlation matrices of the interference and cascaded channels is assumed. The mean-square-error of channel estimates is computed and used to optimize the phase-shift configurations of RIS elements. Numerical results prove the effectiveness of the proposed approach in directional or isotropic channel and interference scenarios.