Complex sulfide ores are difficult to process and often require multi-stage sequential flotation. Process outputs such as grade and recovery in each stage are affected by various sub-processes in the system, and it is crucial to monitor the performance in order to maximize the production. In this work, we have proposed and implemented a dynamic monitoring scheme using fundamental modeling and an online soft sensor network for real-time measurements of grade and recovery. Dynamic fundamental models for lead and zinc recovery were developed to represent the multi-stage rougher flotation for lead-zinc sulfide ores. A soft sensor network was built to measure the grade and recovery in real-time using support vector machine classification and regression on multivariate image data. A factorial design with feed particle size, collector dosage in the lead rougher flotation stage, and collector dosage in the zinc rougher flotation stage as the design variables was used to obtain diverse process conditions for validation. Successful validation at the entire range of process conditions demonstrates the potential of the technique for use in process control and monitoring applications. Changes in the collector dosage were monitored in the lead and zinc rougher flotation stages using state and parameter estimates of the fundamental model structure. The process monitoring framework can be extended to monitor other key variables in the process.