In this work, we assessed the spectral analysis method as a local parameter real variance estimator in the conventional Monte Carlo (MC) and the partial-current based coarse-mesh finite diffusion (p-CMFD) method assisted MC. Due to the inter-cycle correlation, the sample variance is largely under-estimated in MC eigenvalue calculations with the conventional single batch run. Recent studies on local tally real variance estimation inspired from the time series analysis suggested the spectral analysis as an excellent alternative. However, the spectral analysis method inherently has a variance-bias trade-off issue in conventional MC. Based on two numerical tests, a 1-D infinite homogeneous reactor and a 2-D SMR problems using the ENDF/B-IIV.1 library, the p-CMFD feedback is confirmed to resolve this issue by effectively mitigating the inter-cycle correlation.