In pole-placement control, offset removal techniques against unknown disturbances containing model errors are classified into two methods: (I) disturbance compensation method, and (II) integration method. Most of the previous studies recommend method (II). In this paper, the author compares both methods and shows that method (I) is comprehensive, including method (II) as the simplest case. Compared with method (II), method (I) can realize smaller input/output variances and larger robustness to model errors by designing the filter for disturbances (the disturbance observer). Hence, an optimal design procedure for the filter is proposed to take into account the estimated magnitudes and the characteristics of disturbances and model errors. A numerical simulation illustrates the advantages of method (I).