Statistical white reflectivity, minimum-phase wavelets, and stationarity are regarded to be three major shortcomings in dealing with the Robinson convolution model. Modern reflectivity inversion methods (e.g., constrained reflectivity inversion) mainly attempt to reduce side effects of the first two assumptions, but most of them overlook the important fact that seismic signals are typically nonstationary. Numerical tests and practical applications have verified that nonstationarity does have great influence on reflectivity estimation and probably would result in misplaced positions or wrongly restored amplitudes for estimated reflectivity. Nonstationarity can be tackled from a perspective of hyperbolic smoothing in Gabor deconvolution. Specially, the energy relationship of hyperbolic strips in a log spectrum can be taken as a quantitative indicator in balancing nonstationarity and conditioning seismic traces to the assumption of unchanging wavelets. Applications to marine seismic data show that balancing nonstationarity finally helps to recover subtle reflectivity information and promotes better displays of geologic features of the subsurface.