Compositional layering of intrusive rocks is often cited as a source of seismic reflectivity in the crystalline crust. Direct evidence for this is based primarily on synthetic seismograms calculated from laboratory measurements of rock velocity and density, or estimates of aggregate rock properties from measurements of mineral velocities and densities. Little is known about how in situ effects such as fracturing, chemical alteration, or anisotropy, may change the reflectivity character of the surface seismic data. An ideal dataset for examining the hazards of using physical properties data for estimating crustal reflectivity occurs at the Nellie intrusion, a middle Proterozoic-age layered mafic intrusion that lies beneath the Permian Basin of west Texas. Due to a well that penetrated 4.5 km of the intrusion, we have a dataset that spans the scale from compositional data on the intrusion derived from well cuttings, to log data, to surface reflection data. Sonic and density log data from the well measure velocities and densities 10 to 15% below the values calculated from modal mineralogy. We attribute this difference primarily to pervasive post-emplacement alteration and fracturing of the rock mass in situ. Thus, this unique dataset provides a quantitative estimate of the magnitude of difference between observed and theoretical velocities that can be expected for crystalline rocks in the upper crust and its cause. Despite the difference, analysis of synthetic seismograms shows that primary compositional variation of the intrusion is still responsible for sub-horizontal layered reflectivity in both the log-based and petrology-based synthetic seismograms. Boundaries of intrusive cycles are characterized by large negative reflection coefficients due to the juxtaposition of mafic-rich basal layers against plagioclase-rich layer tops. In detail, only the log-based synthetic provides a good match to the surface seismic reflection data. This suggests that physical properties from modal mineralogy are useful in obtaining a generalized model for reflectivity, but that differences in bulk physical properties of the rock mass preclude the possibility of obtaining a one-to-one match with surface seismic data.