One key issue for the successful deployment of OpenGIS technology is the design of application schema that can correctly model the spatio-temporal nature of reality phonomena. Such a design must be able to (1) record the complex spatio-temporal status of selected features, (2) encode all required information in a consistent architecture and (3) reduce unnecessary duplicated contents as much as possible. By first analyzing the essential characteristics of geospatial data, we argued the modeling of individual feature should at least include six components: identifiers, time, spatial description, positional accuracy, attribute and attribute quality. After further exploring the possible scenario of recording data, four distinct primitive feature types that capable of modeling simple to complex spatio-temporal phenomena are summarized. Each feature type includes all of the six above components, but has its own inner structure to avoid unnecessary duplicated data contents. To simplify the selection of referenced feature type, a decision tree is developed to provide guidance to the designer. The proposed feature architecture is implanted in GML to allow open and interoperable interpretation. Under such circumstances, features referred to the same primitive feature type can be parsed in the same way to reduce the cost of application development. Furthermore, the spatial, temporal and quality status of all selected geospatial features can be unambiguously acquired. To make all data available in open format is certainly preferred, further standardization toward the modeling architecture would reduce the complexity to both designer and application developers.