This paper shows that an extremely accurate 333 interpolator can be designed in the frequency domain. The design is a constrained weighted least-squares (WLS) problem. The 333 interpolation kernel consists of the third-order piecewise polynomials with 12 parameters. Subject to some design constraints, the remaining unconstrained (free) parameters can be optimized to minimize the zero-phase frequency-response (FreRes) error in the frequency band of interest. In this paper, we first detail the general form of FreRes of the 333 interpolator, and then derive the FreRes expression using free parameters. Finally, we show how to optimize the free parameters by minimizing the squared FreRes error. The design formulation only considers the squared FreRes error and ignores the FreRes error in unimportant frequency bands. This produces an extremely accurate interpolator in the frequency band of interest.