A robust epistemology is a sufficient condition for a successful pursuit of scientific inquiry. There are many other factors and conditions that must be met for science but a vigorous epistemic model for how one pursues scientific inquiry is needed; otherwise, there may be sufficient reasons to doubt not only the conclusions of the scientific inquiry but as well as the pool of data, which must be assessed appropriately. The scientist is more than welcome to pursue an empiricist model for his epistemology, though strict [naturalistic] empiricism is not very robust, but it must have certain allowances for metaphysical import—perhaps more rationalistic.
I believe the best way to construct a robust epistemology and scientific method is to be a realist. What I mean by this is that the external reality is how it appears to be to an observer making an epistemic inquiry, the measurements from science accurately depicts reality. This is in contrast to instrumentalism/anti-realism, which suggests that our inquiry of the world, scientifically, do not accurately depict reality but as useful fictions. An instrumentalist is more concerned about data fitting theories and predictions than with an accurate depiction of reality.
For the scientific realist, the ontology of the world determines one’s epistemology. They congruently correspond. It is important to note the order of entailment. Antecedently, reality determines our epistemology. It would be illicit to reverse the term order would be an epistemic fallacy. I am not advocating a naïve realism where reality acts on the human mind without personal inquiry nor am I advocating postmodern anti-realism where one can construct whatever type of reality is desired. I am advocating a form of critical realism (or critical non-realism in certain instances concerning theoretical entities when there may be reason to suspect insufficient knowledge of the theoretical entity).
The first issue concerning an empiricist epistemology would be the issue of justification. The justification of our beliefs in propositions describing physical objects is always inferential and that it is always from propositions about the nature of our experiences that such inferences are made. With this said, there are two conditions that must be satisfied concerning inferential belief in physical objects: 1) Statements about experience must count as reasons or evidence for statements about objects and 2) Statements about experience must in some, no doubt rather obscure, sense be accepted by those who make statements about objects. However, in order to ground these beliefs one must always be in the appropriate phenomenological frame of mind. All conditions for a right phenomenological state of mind aside, one must account for inferential beliefs. Some beliefs are inferentially justified but are not usually if ever consciously inferred from that which would constitute evidence, which would be beliefs about the future or subjunctive conditionals. The belief that this glass of water in front of me will quench my thirst if I drink it is not inferred back from previous experiences coupled with an application of a synthetic a priori principle of induction. Though this example is not how we form our beliefs psychologically or historically, it can be formed via instances of past experience and induction in the logical sense.
The reason why inferential beliefs are so important is that one’s scientific method cannot be contrary to one’s epistemic method. With that said, certain models for scientific explanation must have justificatory acceptance. For example, a deductive form of scientific inquiry cannot be the only means acceptable since one cannot have a deductive form of epistemology since all beliefs would be self-justified and self-preserved (at least this would not account for a robust epistemology).
Such methods are derived from the use of abductive reasoning. The American philosopher and logician Charles Sanders Peirce first described abduction. He noted that, unlike inductive reasoning, in which a universal law or principle is established from repeated observations of the same phenomena, and unlike deductive reasoning, in which a particular fact is deduced by applying a general law to another particular fact or case, abductive reasoning infers unseen facts, events, or causes in the past from clues or facts in the present.
Certain unknown entities may become known by inferential means. We can infer the existence of protons, quarks, and other elementary particles by predicting what effects such entities may have in certain situations. This may be causal in nature and may be confirmed by inference. However, it is not the case that we directly experience the existence of these particles (for all intents and purposes, it certainly is the case that we experience particles when we run in to a wall and even then we experience the strong nuclear force over the particles). There is no direct evidence for the existence of these theoretical entities. Additionally, theory cannot discharge its explanatory function without them (though this serves no problem for anti-realism).
Nevertheless, epistemological direct realism and new belief formation can be non-inferentially justified. The reason why non-inferentially justified beliefs must also be accepted is that such beliefs may be basic from which we may make further inferences. Other propositional beliefs may be basic but non-empirical such as mathematical truths. What is interesting about having mathematical truths being non-inferentially justified and basic is that mathematics seems to be the language in which physical phenomena and laws are expressed. If it’s true that physics and the empirical world are expressions of mathematics, Platonism aside, then it seems that the most fundamental aspect of the empirical world is not empirical itself.
When using certain theoretical terms, as in the inference to quarks, the epistemic process cannot restrict explanations to only natural or empirical explanations. If one attempts to strip science of all metaphysical import then material causation is the only sufficient form of scientific explanation. However, this has an unnecessary restriction on science and is incongruent with one’s epistemology (if it is to be robust). The robust epistemology certainly accounts for inferential explanations that are not necessarily required to be material. The epistemic methodology may be identical to a non-scientific context but when this methodology is applied in a scientific context then the explanation is ruled out a priori with no [apparent] justification (hence the removal of efficient and final causation from science). Thus, scientific explanations must not necessarily be material explanations. Remember, by using inferential explanations such as quarks and protons we observer their effects and infer as to what the best antecedent causal explanation may be. It’s an issue over the identity of what antecedent causes could be. (In a normal epistemic process the antecedent may be agency). Being an epistemological and scientific realist does not exclude one from being a metaphysical realist.
 R.A. Fumerton, “Inferential Justification and Empiricism.” The Journal of Philosophy 73 (1976): 559-60.
 Charles Sanders Pierce, “Abduction and Induction” in The Philosophy of Pierce, ed. J. Buchler (London: Routledge, 1956), 375.
 Alex Rosenberg, Philosophy of Science (New York: Routledge, 2012), 158.