This paper presents a new learning model called the statistical query model, which is weaker than Valiant's original learning model but allows for efficient noise-tolerant learning. The statistical query model replaces Valiant's example oracle with a statistical query oracle that provides estimates of probabilities over the sample space. The paper shows that any class efficiently learnable from statistical queries is also efficiently learnable in the presence of classification noise. This result is demonstrated through several applications, including efficient noise-tolerant learning algorithms for conjunctions, perceptrons, axis-aligned rectangles, and AC⁰ circuits. The paper also investigates the relationship between the statistical query model and the classification noise model, showing that they are not equivalent. The paper concludes with open problems and further research directions.This paper presents a new learning model called the statistical query model, which is weaker than Valiant's original learning model but allows for efficient noise-tolerant learning. The statistical query model replaces Valiant's example oracle with a statistical query oracle that provides estimates of probabilities over the sample space. The paper shows that any class efficiently learnable from statistical queries is also efficiently learnable in the presence of classification noise. This result is demonstrated through several applications, including efficient noise-tolerant learning algorithms for conjunctions, perceptrons, axis-aligned rectangles, and AC⁰ circuits. The paper also investigates the relationship between the statistical query model and the classification noise model, showing that they are not equivalent. The paper concludes with open problems and further research directions.