In this paper, a comparative study of the two methods, the multiple-models and just-in-time learning (JITL) method, for empirical modeling of a pilot pule-jet fabric filter is presented. A salient feature of this filtration process is that its dynamics frequently undergoes changes caused by the variation in the dust concentration of the inlet air stream. Experimental results show that JITL not only gives better overall modeling accuracy but also it can predict the filter dynamics in the absence of the inlet dust concentration, which is either imprecisely known or unavailable in the industrial applications.