We examine uncertainty in predicted linear 1D site amplification due to uncertainty in shear-wave velocity () structure quantified from Bayesian (probabilistic) inversion of microtremor array dispersion data. Based on a sample of profiles drawn from the posterior probability density of the microtremor inversion, probability distributions are computed for common predictors of site amplification including (traveltime average to a depth ) and amplification spectra based on seismic impedance variations and full transverse shear-wave effects. These methods are applicable for any site, but the resulting probabilistic site amplification analyses are specific to the two sediment sites studied here with strongly contrasting geology in high population centers of British Columbia, Canada. The site amplification probability distributions for the two sites are shown to be more informative than amplification estimated for a single best-fit profile by characterizing the uncertainty and therefore level of confidence in the predictions. The shear-wave amplification probability spectra are evaluated by comparison to empirical earthquake and microtremor spectral ratios, with generally good agreement in resonant peak frequencies and amplification levels, providing confidence that the primary influence of site-specific structure is accounted for appropriately. The wider implication here is that proper characterization of the profile uncertainty distribution from inversion of cost-effective surface wave dispersion data is beneficial in the application of said profiles to the prediction of earthquake site response and its uncertainty, as required for probabilistic seismic hazard assessment.