The Serial Reaction Time task is one of the most commonly used tasks to measure procedural learning. Despite its frequent use both in children and adults, this task has been found to show sub-optimal test-retest reliability when compared to clinical measures, whose reliability coefficients are usually above .80 (Portney & Watkins, 2015). This pattern is especially noticeable in children (West et al., 2019). Since test-retest reliability refers to the ability of the test to yield similar results for repeated assessments of the participants behaviour, ensuring the stability of the measuring instrument is crucial as neither researchers nor clinicians can draw conclusions from inconsistent performances. In cases where the reliability is very poor, as has been found in the SRT task when used in children (West et al., 2018; 2019), inferences based on this task will be mostly unreliable (Wagenmakers & Gronau, 2017). Thus, efforts have been put forward to determine the best practices to address the stability issues of the Serial Reaction Time task. Study 1 (N = 103) investigated whether the similarity between sequences used during Session 1 and Session 2 would influence the amount and/or stability of procedural learning. The results demonstrated clear evidence of procedural learning effects in both sessions, as well as a positive relationship with similarity: more similar sequences in Session 2 showed larger learning effects than less similar sequences. In terms of reliability, although split-half reliability was acceptable (r > .55), test-retest reliability was poor, regardless of similarity (r < .25). Given the low levels of test retest reliability, Study 2 went on to replicate the design used by West et al (2019) which reports the highest levels of stability to our knowledge, Study 2 (N = 47, age range 17-34; mean = 20.09, SD = 2.09) went on to replicate the design used by West et al. (2019) as it had produced the highest levels of stability. Additionally, this study aimed to test whether the test-retest of the SRTT would increase over multiple sessions (Palmer et al., 2018), and whether the task would be sensitive to individual differences in language and literacy (as predicted by the Procedural Deficit Hypothesis (Ullman, 2004)), and sustained attention (in accordance with West et al. (2019)). As before, participants consistently showed evidence of procedural learning in all sessions, with practice effects decreasing from session 2 to 3. Split-half reliability was higher than test-retest reliability and stability of the SRT task was higher between sessions 2 and 3 than between sessions 1 and 2 (test-retest r = .60 and r = .43, respectively). Even though the test-retest reliability was considerably higher than for Study 1, particularly for the last sessions, it was still lower than the stability coefficient obtained by West et al. (2019; r = .66) after just two sessions, especially when using the same procedural learning measures, i.e. difference scores. The lower stability observed in Study 2 relative to West et al. (2019) may potentially be explained by two different design features: the age of the participants and the presence/absence of an interstimulus interval. Firstly, West et al. (2019) analysed the performance of older participants (18 - 61 years, mean = 25.33 years; SD = 10.33 years) than those who participated in Study 2 (17 - 34 years, mean = 20.09 years, SD = 2.09 years). This could have contributed to increasing the stability of the SRT task as test-retest reliability has been found to increase with age in intelligence measures (Schuerger & Witt, 1989). Secondly, West et al. (2019) included a 250 ms interstimulus interval (ISI) between trials which was absent in our previous experiment in an attempt to reduce explicit awareness (Destrebecqz & Cleeremans, 2001; 2003). The inclusion of an ISI, however, could have contributed to the higher test-retest reliability by inducing better quality representations (Cleeremans & Sarrazin, 2007; Gaillard, Destrebecqz, Michiels, & Cleeremans, 2009), with explicit awareness possibly emerging as a consequence of the gradual increases in strength and stability of the representations (Cleeremans, 2011; Timmermans et al., 2012). Thus, to determine whether these factor(s) contributed to the lower stability in our previous experiment, this study will compare the performance of participants in the SRT task in two separate groups, one group will perform the SRT task with an ISI of 250 ms, as in West et al. (2019), and the other will perform the task without an ISI, as in experiment 2. To further explore the effect of age on the stability of procedural learning, we will recruit participants in the age range of those included in West et al. (2019), with groups being age-matched to avoid age related differences between groups. Thus, it will be possible to compare the test-retest reliability of the two groups with ISI and no-ISI; as well as verify if there are age-related changes in test-retest reliability by dividing groups in younger and older adults. Another purpose of the present study is to assess the reliability of an online version of the SRT task as a reliable online measure may hold practical advantages for research and clinical purposes. Not only would it allow for mass testing of populations in naturalistic settings, but would also have the potential to increase reach to a more diverse array of participants and make it more likely that the sample will be representative of the population, relative to testing undergraduate populations (Raz et al., 2014). Finally, as predicted, Study 2 replicated the positive and significant correlation between procedural learning and attention observed by West et al. (2019). The relationship was particularly strong in sessions 1 (r = .44) and 2 (r = .49), with a decrease in session 3 (r = .25). This pattern could be potentially explained by Thomas et al. (2004) findings which showed less reliance on attentional resources once the sequence becomes more predictable since, even though distinct sequences are used per session, their level of similarity is still high. Thus, comparing the relationship between procedural learning and attention in both groups may provide a better understanding of the role of attention in the SRT task, as the predictability of the sequence would be expected to be higher in the group with ISI (Destrebecqz and Cleeremans, 2001; 2003).