The flight was assigned an airspeed of 230 knots and cleared to descend from FL240 to 16,000 feet in preparation for landing at Miami. The FDR indicated that while the autopilot was engaged in the descent, the power levers moved from the mechanical autothrottle limit of 44 degrees to the manual limit of 37 degrees. As the aircraft leveled at 16,000 feet the airspeed decreased. The F/O began a right turn to enter a holding pattern and added some power, which stabilized the airspeed at 178 knots. However, the right bank and the resultant angle of attack (AOA) continued to increase, despite left aileron input by the autopilot. As the autopilot reached the maximum input of 20 degrees, bank angle increased past 50 degrees, and the AOA increased rapidly from 7 degrees to 12 degrees. At this point the stick shaker activated, the autopilot independently disconnected, the power was increased, and full left rudder was used to arrest the roll. The bank angle reached 56 degrees, and the AOA reached 13.7 degrees at 177 knots. The aircraft then pitched down, and entered a series of pitch, yaw, and roll maneuvers as the flight controls went through a period of oscillations for about 34 seconds. The maneuvers finally dampened and the crew recovered at approximately 13,000 feet. One passenger was seriously injured and one flight attendant received minor injuries during the upset. According to wind tunnel and flight test data the A300 engineering simulator should adequately represent the aircraft up to 9 degrees AOA. Unlike the accident aircraft; however, the simulator recovered to wings level promptly when the lateral control inputs recorded by the FDR were used. The roll disagreement between the simulator and accident aircraft began at 7 degrees AOA, and it appears that some effect not modeled in the simulator produced the roll discrepancy. Just prior to the upset the accident aircraft entered a cloud deck. The winds were approximately 240 degrees, 35 knots, and the ambient air temperature was approximately minus 4 degrees C. An atmospheric disturbance or asymmetric ice contamination were two possible explanations considered, but unproven.