The idea of constructing a data-driven stochastic system model through subspace identification for the purpose of inferential control is investigated. Various available methods for designing an inferential controller are discussed and their limitations are brought out, particularly in applications involving multi-variable processes. Practical issues that arise in identifying a system model geared toward inferential control using a subspace method are discussed. They include: handling of nonstationary disturbances, handling of multi-rate measurements/missing data, and secondary measurement selection. With the identified stochastic system model, a multi-rate Kalman filter can be designed and coupled with a model predictive controller. The method is applied to a continuous pulp digester, which is a complex distributed parameter system involving heterogeneous reactions. The application study indicates much potential for the data-based approach.