We use a large dense array of seismometers to quantify spatial variation in site response, and its effects on estimates of earthquake stress drop, and also how source complexity leads to bias in spectral source modeling. Stress release during an earthquake is proportional to the slip divided by the length scale of the rupture. It is a basic property of earthquakes fundamental to understanding the physics of the source and its energy budget. The stress release also governs the amplitude of ground motions at the frequencies important for strong ground motion prediction and so is inherent to seismic hazard analysis. Many studies have attempted to characterize the high-frequency earthquake radiation spectrum and measured stress release using simple source models, to distinguish induced and tectonic earthquakes, and determine the factors controlling the earthquake rupture process. Small and moderate sized earthquakes are typically used for studies of source characterization because of their larger numbers, especially in regions of lower strain rate. Unfortunately, this work has led to inconclusive and controversial results, with different studies failing to agree within their calculated uncertainties. Ongoing analysis has found the major sources of error result from distinguishing source from path and site effects, and assuming simplistic source models, both a consequence of the limited data for most events, in terms of number of stations and frequency range of the signal. Large-N arrays of seismometers provide an unprecedented opportunity to characterize smaller earthquake sources, using extremely well recorded wavefields without spatial aliasing, and quantify the real uncertainties from analyses using smaller station numbers and various methods. The LArge-n Seismic Survey in Oklahoma (LASSO), a USGS deployment in 2016, is the largest public archive to date (in terms of spatial coverage: over 1800 stations, and frequency range: 500 samples/s) to record the wavefield of smaller earthquakes. We investigate the site response across the dense array, by calculating relative peak ground velocity from regional and teleseismic events in different frequency ranges. We find a strong correlation between the site amplification at high frequencies and the surficial geology. Sites with high amplification are typically located on young alluvial sedimentary deposits. At lower frequencies the PGV is more dependent on the earthquake radiation pattern. We then use the site amplification effects and regression analysis to correct the measurements of seismic moment and corner frequency previously obtained using standard spectral fitting. We find that the estimated site effects decrease the spatial variability of the source parameters significantly. Finally, we apply empirical Green’s function methods in both the frequency and time domains to quantify the uncertainties in estimates of stress release and finite rupture extent of M2-3 earthquakes. We find that source complexity is a major cause of uncertainty, because it decreases the appropriateness of the simple source models in common use. We consider ways in which the variability resulting from source complexity can be accurately included in analysis of more typically-recorded events.