Rod-pumping is the most common artificial lift method used to move oil to the surface in low pressure resevoirs. A variety of mechanical problems can occur with this system. The adjustment of the pumping capacity to the reservoir production rate is another source of error compromising pumping efficiency. The need to identify these problems quickly and accurately is essential in attempts to minimize operating costs and maximize production. A dynamometer is attached in general to the polished rod of the pumping unit, and a plot of load vs. position, known as (downhole) dynamometer card (DC), is obtained for the purpose of rod load monitoring. Altough DC is a very important piece of information, other data may be required to support complex decision making about the actual rod-pumping condition. For this purpose, the engineer uses also information about the characteristics of the well, the type of oil being pumped, etc., besides taking into consideration the DC shape, maximum and minimum load values, etc. A new type of neuron is used to built neural nets having powerful numeric and symbolic processing capabilities, besides permiting knowledge to be encoded not only on the wiring of the net, but also on the selection of the types of neurons and synapsis composing the net. This new type of neural nets was used to develop SICAD, a hierarchical neural system whose purpose is the intelligent control of rodpumping. SICAD is composed by two famillies of neural nets specialized, respectively, in pattern recognition (PRN) and expert reasoning (ERN). Different modes of interactions 97 between ERN and PRN define different pumping control strategies.