This study tested whether different referencing and data preprocessing strategies would influence the results of electroencephalographic network analyses. Data were collected from two volunteers during the last 60 seconds of the second and fourth rapid-eye-movement epochs using a 256-channel electroencephalographic system and prior to correlation analyses, were preprocessed by varying combinations of techniques, including referencing/re-referencing (vertex, average mastoid, and common average), bandpass filtering, differencing, and autoregressive integrative moving average (ARIMA) modeling. The findings suggest that using different methods to eliminate temporal structures, noise, and extraneous signals embedded in a time series does not substantially influence the results of subsequent network analyses. However, ARIMA parameters should be carefully chosen to ensure that all or almost all 256 EEG time series can be preserved after the ARIMA prewhitening procedure. In contrast to the analogous correlation networks between different data preprocessing protocols, different referencing can lead to significantly dissimilar correlation patterns. Specifically, the use of a common average reference moderates the overall strength of a connectivity network, while vertex referencing and mastoid referencing may respectively underestimate the strength of connections over the central and mastoid regions. All in all, it appears that the combination of average referencing, filtering, and differencing provides a relatively reliable protocol for dense EEG network analysis.