Rand left the exposition of her theory of perception to Leonard Peikoff and David Kelley. It is summed up on the Stanford Encyclopedia of Philosophy as follows:
Our perceptual faculties place us in direct contact with reality. In this sense Rand's theory of perception is a version of direct realism, holding that the objects of perception are extramental entities (rather than, say, subjective experiences on the basis of which we infer entities as their causes)....
Rand rejects the view that some perceptions are of the qualities of objects as they are independently of us (primary qualities), whereas others (secondary qualities) are caused by the primary qualities, and are entirely in the mind. Instead, she distinguishes between the content of a perception and its form; when we perceive an object as, e.g., square and red, what we perceive are its intrinsic features in a certain form, a form that is determined by the nature of the object, the nature of our perceptual organs, and the environment. Thus, we perceive the object's shape as square, and the reflectance properties of its surface as red; both are the result of the interaction of our perceptual organs with what is out there. Neither squareness nor redness belong either to the object apart from our mode of perception, or to our mode of perception apart from the object in its environment. Hence, these attributes are neither intrinsic nor subjective but relational and objective.
Thus while Rand is a direct realist in the sense explained above, she is not a naive realist in the sense of regarding all perceived attributes as enjoying equal extramental status.
The point in dispute involves the term "extramental." As far as I know, that is not a term either Rand or Peikoff have ever used. Its use is required only because Objectivists have never told us whether percepts are mental or not. That is an issue they have artfully dodged, and here is why: if they were to admit that percepts are mental entities (as they admit, for example, that concepts are), they would be confessing that their realism is indirect and perhaps even representational. So when the Stanford Encylopedia decided to do a write-up on Ayn Rand (under the advisement of such Objectivist worthies as David Kelley and Gregory Salmieri), they had no choice but to describe percepts as "extra-mental."
I suppose orthodox Objectivists might have prefered the term "objective," even though it doesn't apply in this context. Yes, fine, percepts are "objective." But that doesn't answer the question as to their ontological status. Now Objectivists have no problem describing concepts as both objective and as "mental entities." So what's the issue with percepts? Why can't a percept be every bit as mental as a concept? Rand defines a percept as "a group of sensations automatically retained and integrated by the brain of a living organism." [IOTE, 5] Very well: so why can't groups of sensations integrated by the brain be regarded as "mental"? Percepts, after all, are only experienced in the mind. Percepts don't exist in matter. An apple becomes a percept only if it is perceived by a mind. So why isn't it mental?
The challenge is fundamental. We get signals from the external world through our sense organs. When we experience these signals within consciousness, why should we not classify them as mental? What compelling reason is there for denying their mental status?
Rand's main objection to regarding percepts as mental is that it leads to indirect realism and (even worse) "representationalism." This form of realism, Rand believed, cuts the mind off from reality. Representationalism contends that we have "direct" access only to sense data (i.e., percepts); and that we subsequently infer external reality from the experience of this mental data. One problem emerges: how is this inference justified? How do we know that the sense data experienced within consciousness actually corresponds to something outside of consciousness? For that matter, how do we know that there's anything out there at all, if all that we experience (directly) is sense data?
Rand seems to have believed that these questions could be answered by merely denying that perception is mental. Instead, she argued, perception is the "form" through which the mind experiences the content of reality. But why isn't this form through which the mind perceives reality itself a mental entity? Representationalists talk about perceiving reality through a screen or a veil of ideas. Objectivism is not much different. Instead of screen of ideas, we have a screen of forms. Rand tried to solve this issue by juggling terms. However, it's not clear in what respect the Objectivist view differs from the representational/indirect view except in terms of rhetoric.
If it is argued that Rand's position is different because she stresses the process by which these forms are grasped by the mind, then I would reply: there's no reason why an indirect realist can't place the same emphasis on process. Indeed, the very inference on which indirect realism is based presupposes that there is a process involved by which the mind attains knowledge about the external material world. Representationalists contend that we infer the objects of reality from internal sense data. But this gets it backwards. Rather, we infer representationalism from the fact that we perceive external objects through perceptions. That is to say, the assumption of realism precedes, and almost certainly must precede, the assumption of representationalism. When the adventure of knowledge begins, way back in infancy, no inference from sense data is even possible. For it is grossly implausible to assume that, when the infant looks at his mother, he first thinks, that is a sense data and only subsequently infers his mother from his perceptions of her. On the contrary, all the evidence we have strongly suggests that we begin as direct realists. The sense data is assumed, by a kind of biological prompting or instinct, to convey direct information about the external world right from the start. It is only very later, when (or rather if) we become discriminating philosophers, that we realize that our earlier, instinctive realism has been naive, and that we must revise it to more accurately convey what is in fact happening. If we study perception under realist assumptions, we discover a process by which the mind interprets the data of sense and forges the various percepts or sense data (call them what you will) that flame across the individual's consciousness. Are these percepts or sense data in fact mental? How could they not be? Part of the process by which the mind discovers pertinent facts about the external world involves experiencing sense data. Realism is therefore indirect: the external world is experienced via a process that involves acquaintance of sense data. Knowledge is symbolic, rather than direct or literal.
On realist assumptions, no other sort of knowledge is even possible. A "direct" knowledge would be merely knowledge of sense data, which is not knowledge at all, but merely reverie and delusion. Knowledge only arises when our sense data is taken to be a signal or report of things, entities, processes, attributes, properties existing outside of consciousness. If the mind is not intelligent enough to make this distinction, it will never know anything beyond its own dreams.