SummaryIn reliable orthogonal frequency division multiplexing systems, channel estimator (CE) plays a crucial role. A pilot‐aided orthogonal frequency division multiplexing CE often employs denoising thresholds to identify significant and noisy channel impulse response taps. When the knowledge of channel statistics (KCS) is completely unavailable, the existing suboptimal thresholds require consistent estimation of one or more KCS parameters. In this paper, the effect of noise variance parameter estimation on CE's mean square error (MSE) performance is characterized analytically by computing threshold's tap‐detection probability. To mitigate MSE degradation due to KCS estimation, an eigen‐select threshold, which does not require any KCS estimation, is proposed. It utilizes eigen values of the auto‐covariance matrix formed by estimated channel impulse response. The proposed threshold is compared with existing state‐of‐art alternatives, for CE's MSE performance and system's BER performance, in correlated Rayleigh fading channel environments. Results reveal that, with the proposed threshold, performance plots converge to optimal solution more closely than the others. Copyright © 2015 John Wiley & Sons, Ltd.