Summary: QR-decomposition-based least-squares lattice (QRD-LSL) predictors have been extensively applied in the design of order-recursive adaptive filters. This work presents QRD-LSL interpolators that use both past and future data samples to estimate the present data sample based on a novel decoupling property. We show that the order-recursive QRD-LSL interpolators display better numerical properties and achieve a higher level of computational efficiency than the conventional LSL interpolators. Except for an overall delay needed for physical realization, QRD-LSL interpolators can substantially outperform QRD-LSL predictors while retaining all the desirable features of the latter.