Mechanical properties of engineering materials are sensitive to the underlying random microstructure. Quantification of mechanical property variability induced by microstructure variation is essential for the prediction of extreme properties and microstructure-sensitive design of materials. Recent advances in high throughput characterization of polycrystalline microstructures have resulted in huge data sets of microstructural descriptors and image snapshots. To utilize these large scale experimental data for computing the resulting variability of macroscopic properties, appropriate mathematical representation of microstructures is needed. By exploring the space containing all admissible microstructures that are statistically similar to the available data, one can estimate the distribution/envelope of possible properties by employing efficient stochastic simulation methodologies along with robust physics-based deterministic simulators. The focus of this thesis is on the construction of lowdimensional representations of random microstructures and the development of efficient physics-based simulators for polycrystalline materials. By adopting appropriate stochastic methods, such as Monte Carlo and Adaptive Sparse Grid Collocation methods, the variability of microstructure-sensitive properties of polycrystalline materials is investigated. The primary outcomes of this thesis include: • Development of data-driven reduced-order representations of microstruc- ture variations to construct the admissible space of random polycrystalline microstructures. • Development of accurate and efficient physics-based simulators for the estimation of material properties based on mesoscale microstructures. • Investigating property variability of polycrystalline materials using efficient stochastic simulation methods in combination with the above two developments. The uncertainty quantification framework developed in this work integrates information science and materials science, and provides a new outlook to multiscale materials modeling accounting for microstructure and process uncertainties. Predictive materials modeling will accelerate the development of new materials and processes for critical applications in industry.