Electroencephalography (EEG) is an accessible technique that records neuronal oscillatory activity from the scalp. Functional connectivity (FC) analysis through different metrics can be applied to EEG signals to obtain an in-depth understanding of the brain mechanisms. The resulting functional connectivity network (FCN) is of great interest because it has the potential to serve as a biomarker in brain diseases. However, one unsolved issue is that volume conduction (VC) in the EEG system may lead to spurious connectivity and yield inconsistent FCNs across studies. In order to interpret FCNs with confidence, there is a need to develop connectivity metrics that are immune to VC. VC is the spreading of the electric field through the scalp collected by multiple electrodes simultaneously. Multivariate metrics have been reported to be insensitive to VC by means of removing the effect of confounding variables when it quantifies the relationship between two variables. Here, we propose a new multivariate metrics Partial Cross-Correlation (PCC), aiming to reduce the effect of VC in the FCNs. We assessed partial correlation (PC) and PCC as multivariate metrics to construct FCNs by applying these metrics to simulated data and human EEG. The simulated data were generated by the Kuramoto model and Rössler model, two models that represent brain dynamics. The human data consisted of routine EEG studies from 240 healthy infants, aged 0-2 years old. The results are compared to FCNs from other well-established bivariate methods, cross-correlation (CC) and adjusted cross-correlation (ACC). Results show that PC, PCC, and ACC excluding values at zero lag (ACC-e) are all valid choices for the EEG system. PC performs the best against volume conduction among all the metrics in both simulated and human EEG. PCC is similarly insensitive to volume conduction, but it is 100-times slower to calculate compared to PC. ACC-e is the only bivariate metric that can work against volume conduction. It is also the most computationally efficient one, yet with the highest false-negative rate. Overall, we recommend using PC or ACC-e as reliable metrics against volume conduction.