The exhaust purification system comprises a control apparatus performing main feedback control controlling the amount of fed fuel so that the output air-fuel ratio of the upstream side sensor becomes a target value, sub feedback control setting the target air-fuel based on the output air-fuel ratio of the downstream side sensor, main learning control controlling the amount of fed fuel based on a main learning value, and sub learning control controlling the amount of fed fuel based on a sub learning value. The control apparatus performs sub learning promotion control so that the sub learning value easily changes to a suitable value when a sub learning promotion condition, which is satisfied when the absolute values of the main learning value and the sub learning value are respectively predetermined reference absolute values or more and these learning values are opposite in sign, is satisfied.