傳統知識論接受推論之證成概念：一信念的證成是藉由其他的信念（或狀態）合理推出該信念。推論證成立即導致「無限後退」的問題，因為具證成能力的信念（或狀態）本身亦需要被證成。本文嘗試提出不會無限後退的推論證成。首先分析傳統推論證成理論具有三原則：狹義推理理論，形式主義的推理概念，以及單線後退的證成方向。此三原則必須被修正，並取代之以新原則：廣義推理理論、實質主義、平衡互動的證成方向。這些原則所組成的新推論證成理論不再有無限後退的問題。

Traditional epistemology embraces an inferential concept of justification: a belief is justified just in case it can be reasonably inferred from other beliefs. Inferential justification immediately incurs the problem of infinite regress, since the justifying beliefs themselves are in need of further justification. In this paper I will try to explicate a new idea of inferential justification that does not regress indefinitely. I will begin by analyzing the traditional notion of inferential justification as involving three principles, namely narrow inferentialism, formalism, and linear regress justification. All three principles are critically examined, and in their place, three new ones are proposed: broad inferentialism, materialism, and interactive justification. The principles constitute a new inferential approach which admits of no infinite regress.