As a sort of partial answer, but since it's too much to add to the question itself... a more encompassing question would be whether fear conditioning can occur without awareness of the conditioning event happening at all. I suppose an affirmative answer to the latter implies the answer is also "yes" to my question because forgetting the association event doesn't seem any more restrictive than not learning (into episodic memory) about it in the first place.
However, the jury seems to still be out on whether unaware conditioning can happen in humans. Some initial (and highly cited) studies answered the question affirmatively, but later studies cast doubt on the results because the level of awareness for the stimuli used seem to vary with the individual. (In other words, there was heterogeneity in awereness in the original studies; for a completely different reason, I've looked into what happens in such a case and asked what the phenomenon is called on cross-validated, but with not very satisfying answers insofar) From a 2005 review by Pessoa, which summarized the research up to then:
Does the processing of emotional stimuli depend upon awareness?
Two influential neuroimaging papers, both of which
appeared in 1998, reported that responses in the human
amygdala occur in the absence of visual awareness.
Awareness was manipulated by employing backward
masking, as introduced by Esteves and Ohman in
a behavioural paradigm. In one study, fearful faces were
shown for 33 ms and were immediately replaced by a
neutral ‘mask’ face that was presented for 167 ms.
Subjects were naive as to the stimulus conditions, which
included masked fearful and masked happy faces. Stronger
responses were observed in response to fearful rather
than happy faces (both masked by neutral faces), even
though, upon subsequent debriefing, subjects did not
report seeing any emotional faces. In another study,
Morris et al. combined backward masking with classical
conditioning to investigate responses to perceived
and non-perceived angry faces. Although the participants
did not report seeing the masked angry stimuli (angry
faces were shown for 30 ms and followed by a neutral
mask that was shown for 45 ms), the contrast of conditioned
and non-conditioned masked angry faces activated
the right amygdala. Recent studies have also gathered
evidence for unaware processing by the amygdala For instance, Glascher and
Adolphs showed that in normal controls, but not in
patients with amygdala lesions, arousal ratings correlated
with skin conductance responses for both subliminal
(unaware) and supraliminal (aware) conditions (see Glossary).
Together with the important earlier work by
Ohman and colleagues, these results have strengthened
the view that emotional processing occurs independently
of conscious awareness. In general, a
stimulus that is shown for about 30 ms before masking
is considered to be at the ‘threshold’ for awareness.
However, a recent study by Phillips et al. did not
observe any response in the amygdala during unaware
conditions. The authors compared aware conditions, in
which target faces (fearful or disgusted) were shown for
170 ms and followed by a mask that was shown for 100 ms,
with unaware conditions, in which target faces were
shown for 30 ms and, again, followed by a mask that
was shown for 100 ms. Unlike prior studies, however,
no differential responses were observed in the amygdala
in association with fearful faces in the unaware condition
(or in the anterior insula for unaware disgusted faces).
A crucial issue in the assessment of awareness is the
criteria used to determine whether a participant is aware
or unaware of a stimulus. According to ‘objective’
criteria, unaware perception occurs when a subject’s
performance in a ‘forced-choice’ task is at chance. Under
such conditions, behavioral effects of unaware stimuli
(e.g. faster reaction time to undetected fearful faces),
as well as the associated fMRI signals, constitute correlates
of unaware perception. According to ‘subjective’
criteria, unaware perception occurs when subjects report
that they are unable to perform the task better than by
chance (independent of their actual objective performance).
Although some of the previous studies used
forced-choice objective methods, they did not assess
performance in a manner that is independent of response
bias on an individual-by-individual basis [...]. This is important because in the face of weak, noisy signals, subjects might indicate that they do not detect
target stimuli and thus appear to be unable to detect them
reliably. A recent behavioral study by Pessoa et al.
addressed these issues by analyzing performance by way
of signal detection methods. The authors varied the
duration (17, 33 and 83 ms) of which a target face (fearful,
happy or neutral) was shown before being immediately
replaced by a neutral-face mask (which was shown for
116, 100, and 50 ms, respectively, such that the target plus
mask lasted 133 ms in each case). The subject’s task was
to report explicitly whether or not a fearful face was
presented in each trial, and also to rate their confidence
in this decision. Although some subjects only reliably
detected the fearful targets that were shown for 83 ms,
36% of the participants were able to detect both the 33 ms
and 83 ms targets. Remarkably, some subjects could even
detect fearful faces that were shown for 17 ms before
masking. These results demonstrated that participants
differ widely in their sensitivity to fearful faces (Figure 1).
Moreover, the results revealed that even very brief
(17 ms) stimuli can be incompletely masked, consistent
with another recent study [Maxwell and Davidson]. The findings by Pessoa
et al. also raise the possibility that the discrepancy
between the results of Phillips et al. nd those of
prior studies might be caused, at least in part, by variability
in sensitivity among individuals, which may be
related to anxiety levels.
I'm not sure if there's a more recent review than this perhaps with different conclusions following newer experiments. The fact that the review was written by someone who also contributed primary research in the "cast doubt" direction casts a little doubt (in my mind) on its objectivity... But the above surely illustrates the difficulty in conducting studies related to awareness.