Reconstruction of neuronal network connectivity from the recordings of neuronal activities is an important problem in neuroscience. In practice, one of the most widely used reconstruction methods is the Granger Causality (GC) analysis. In this talk, I will provide both numerical and theoretical results about the reliability of GC analysis over different sampling interval lengths for the nonlinear neuronal network dynamics. Based on above analysis, two different approaches will be presented for a reliable GC connectivity reconstruction. In addition, I will propose a novel method – Spike Triggered Regression (STR), which employs general features of neuronal dynamics to greatly improve the accuracy of reconstruction. Our STR method yields accurate inference of network connectivity even for a network of dense connectivity or nearly synchronous dynamics, which other reconstruction approaches cannot successfully handle.