We consider a Gaussian stationary process X(t) with an integer analytic spectral density f(λ) and study a problem of its estimation. The process X(t) is non-observable. Instead of it we observe a linear transformation Y(t), 0 ≤ t ≤ T, of X(t) with a transfer function a(A), ∣a(λ)∣=1 if λ belongs to an interval I. We study how far from I the consistent estimation of f(λ) is possible, T → ∞.