Data-derived estimates of water isotope diffusion lengths in ice cores enable corrections for diffusion-induced signal attenuation and are also used to infer past firn temperatures, but rely on fitting the correct spectral model to the observed power spectrum. Established approaches do not account for additional smoothing and aliasing introduced by discrete sampling, which can bias diffusion length estimates. We show that this bias increases with coarser sampling and exceeds 10 % when the sampling interval is greater than approximately twice the diffusion length. By explicitly incorporating sampling effects into the spectral model, we derive an unbiased estimator that improves diffusion length estimates from coarse, on-line measurements and from ice core sections with small diffusion lengths.