Deductive reasoning draws specific conclusions from general premises. When the argument is valid and the premises are true, the conclusion is necessarily true. It is the gold standard of logical reasoning, exemplified by syllogisms ("All mammals are warm-blooded; whales are mammals; therefore whales are warm-blooded") and conditional reasoning ("If it rains, the streets will be wet; it is raining; therefore the streets are wet").
Human Deductive Performance
Humans are imperfect deductive reasoners. Performance on syllogistic reasoning tasks shows systematic biases. The belief bias effect demonstrates that people accept conclusions more readily when they are believable, regardless of logical validity. Atmosphere effects show that people match the form of the conclusion to the form of the premises. Content effects demonstrate that the same logical structure is easier or harder depending on the specific content — Wason's selection task is notoriously difficult in abstract form but easy when framed in terms of social rules (catching cheaters).
Mental Models Theory
Philip Johnson-Laird's mental models theory proposes that people reason by constructing mental models — internal representations of possible states of affairs described by the premises — and drawing conclusions based on what holds across all models. Errors occur when people fail to consider all possible models, typically constructing only one or two models and missing counterexamples that would invalidate a conclusion. The theory successfully predicts which problems are easy (requiring few models) and which are hard (requiring many models).
Dual-process theories (Evans, Stanovich) propose that deductive reasoning involves two systems: a fast, automatic System 1 that generates intuitive responses based on beliefs and associations, and a slow, deliberate System 2 that performs logical analysis. Reasoning errors often occur when System 1's intuitive response conflicts with the logically correct answer and System 2 fails to override it — explaining why people accept believable but invalid conclusions and reject unbelievable but valid ones.