While uncertainty in scientific data attracts an increasing research interest in the visualization community, two critical issues remain insufficiently studied: (1) visualizing the impact of the uncertainty of a data set on its features and (2) interactively exploring 3D or large 2D data sets with uncertainties. In this study, a suite of feature-based techniques is developed to address these issues. First, a framework of feature-level uncertainty visualization is presented to study the uncertainty of the features in scalar and vector data. The uncertainty in the number and locations of features such as sinks or sources of vector fields are referred to as feature-level uncertainty while the uncertainty in the numerical values of the data is referred to as data-level uncertainty. The features of different ensemble members are indentified and correlated. The feature-level uncertainties are expressed as the transitions between corresponding features through new elliptical glyphs. Second, an interactive visualization tool for exploring scalar data with data-level and two types of feature-level uncertainties — contour-level and topology-level uncertainties — is developed. To avoid visual cluttering and occlusion, the uncertainty information is attached to a contour tree instead of being integrated with the visualization of the data. An efficient contour tree-based interface is designed to reduce users’ workload in viewing and analyzing complicated data with uncertainties and to facilitate a quick and accurate selection of prominent contours. This thesis advances the current uncertainty studies with an in-depth investigation of the feature-level uncertainties and an exploration of topology tools for effective and interactive uncertainty visualizations. With quantified representation and interactive capability, feature-based visualization helps people gain new insights into the uncertainties of their data, especially the uncertainties of extracted features which otherwise would remain unknown with the visualization of only data-level uncertainties.