A learning style is a learner's supposed characteristic way of taking in information, with the practical claim that teaching to that channel improves how well the person learns. The popular families sort learners by sensory mode or by a preference profile: VAK (visual, auditory, kinesthetic) and Neil Fleming and Colleen Mills's VARK (1992), which split reading/writing from the visual mode; David Kolb's experiential-learning cycle and its four styles; Rita and Kenneth Dunn's multi-element model; and the Honey and Mumford activist/reflector/theorist/pragmatist types derived from Kolb. The constructs differ wildly in what they claim to measure, but the teaching promise is shared and testable.
The empirically load-bearing claim is the meshing hypothesis: that learning improves when the format of instruction is matched to the learner's diagnosed style, so a "visual learner" learns more from pictures and an "auditory learner" more from spoken delivery. This is the claim that would justify diagnosing styles and tailoring lessons to them, and it is the claim that the evidence has to support for the whole enterprise to earn its place in a classroom. Almost everything sold under the "learning styles" banner rests on it.
A proper test of meshing needs a specific design: diagnose learners' styles, randomly assign learners of each style to different instructional formats, teach everyone the same content, and test everyone with the same test. The signature of a real effect is an interaction, where visual learners do best with visual instruction and auditory learners do best with auditory instruction. Harold Pashler, Mark McDaniel, Doug Rohrer, and Robert Bjork (2008, Psychological Science in the Public Interest 9(3), 105–119) searched the literature for studies using this crossover design and found almost none, and the few credible ones did not show the predicted interaction. Their verdict was blunt: there is no adequate evidence base to justify incorporating learning-styles assessment into general education, and if the practice has utility, that utility remains to be demonstrated.
The instruments themselves are weak. Frank Coffield, David Moseley, Elaine Hall, and Kathryn Ecclestone (2004), reviewing the field for the UK Learning and Skills Research Centre, catalogued 71 distinct learning-style models and examined the leading 13 in depth. Many had poor reliability (a learner retested gets a different style) and poor validity (the score does not predict what it claims to), and several of the most commercially successful instruments fared worst on both counts. A construct measured by an unreliable instrument cannot support a precise matching prescription. Cedar Riener and Daniel Willingham (2010, Change 42(5), 32–35) made the further point that learners do differ, in ability, in background knowledge, in interest, but those real differences are not "styles," and there is no credible evidence that diagnosing a sensory style and teaching to it helps.
What keeps the idea alive is its grip on educators, not its evidence. Philip Newton and Atsushi Miah, and later Newton and Salvi (2020), reviewing surveys across eleven years, found that belief in matching instruction to learning styles ran from 58% to 97.6% of educators sampled, a weighted average near 89%, with no sign of decline despite the accumulating disconfirmation. This persistence is why learning styles is routinely listed among the "neuromyths," intuitive but false beliefs about the brain that survive in education long after research has undercut them.
A fair reading keeps one distinction. People do report stable preferences: a learner may genuinely prefer diagrams to lectures. What collapses under testing is the inference from preference to prescription, the assumption that honoring the preference improves learning. Preference and benefit are different things, and the evidence supports the first while denying the second.
Do not diagnose learners by style or build lessons around a style label, because the matching payoff that would justify the effort has not been shown. Choose the mode the content demands rather than the mode a learner prefers: a map is taught visually because maps are spatial, pronunciation is taught through sound because sound is the point, and no style label changes that. Drawing on multimodal presentation, where words and well-chosen images reinforce the same content, benefits learners in general and is a sounder default than channelling individuals into a single sense. Genuine variation worth responding to is variation in prior knowledge, motivation, and aptitude, which is the proper territory of differentiation and of designing accessible instruction from the start, as in universal design. Keep "learning style" apart from genuine aptitude, a measurable predictor of how readily someone acquires a language, and from multiple intelligences, a separate theory its own author insists is not about styles at all.