The recursive bias compensated least-squares (RBCLS) technique is applied to the estimation of parameters of a linear plant in the presence of input-output and plant input noises. First, we evaluate the bias of the least-squares (LS) estimates of unknown parameters. Then, we derive an RBCLS algorithm that estimates the unknown parameters and noise variances simultaneously with the aid of the technique of Sakai and Arase. Computer simulation studies show that the performance of the RBCLS method is superior to those of RML/RELS and RIV methods applied to the same models.