Purpose The conventional imaging flowchart for prostate cancer (PCa) staging may fail in correctly detecting lymph node metastases (LNM). Pelvic lymph node dissection (PLND) represents the only reliable method, although invasive. A new amino acid PET compound, 18 F-fluciclovine, was recently authorized in suspected PCa recurrence but not yet included in the standard staging work-up of primary PCa. A prospective monocentric study was designed to evaluate 18 F-fluciclovine PET/CT diagnostic performance for preoperative LN staging in primary high-risk PCa. Methods Consecutive patients (pts) with biopsy-proven PCa, standard staging (including 11 Ccholine PET/CT), eligible for PLND, were enrolled to undergo an investigational 18 F-fluciclovine PET/CT. Nodal uptake higher than surrounding background was reported by at least two readers (blinded to 11 Ccholine) using a visual 5-point scale (1–2 probably negative; 4–5 probably positive; 3 equivocal); SUVmax, target-to-background (aorta—A; bone marrow—BM) ratios (TBRs), were also calculated. PET results were validated with PLND. 18 F-fluciclovine PET/CT performance using visual score and semi-quantitative indexes was analyzed both per patient and per LN anatomical region, compared to conventional 11 Ccholine and clinical predictive factors (to note that diagnostic performance of 18 F-fluciclovine was explored for LNM but not examined for intrapelvic or extrapelvic M1 lesions). Results Overall, 94 pts underwent 18 F-fluciclovine PET/CT; 72/94 (77%) high-risk pts were included in the final analyses (22 pts excluded: 8 limited PLND; 3 intermediate-risk; 2 treated with radiotherapy; 4 found to be M1; 5 neoadjuvant hormonal therapy). Median LNM risk by Briganti nomogram was 19%. LNM confirmed on histology was 25% (18/72 pts). Overall, 1671 LN were retrieved; 45/1671 (3%) LNM detected. Per pt, median no. of removed LN was 22 (mean 23 ± 10; range 8–51), of LNM was 2 (mean 3 ± 2; range 1–10). Median LNM size was 5 mm (mean 5 ± 2.5; range 2–10). On patient-based analyses (n = 72), diagnostic performance for LNM resulted significant with 18 F-fluciclovine (AUC 0.66, p 0.04; 50% sensitivity, 81% specificity, 47% PPV, 83% NPV, 74% accuracy), but not with 11 Ccholine (AUC 0.60, p 0.2; 50%, 70%, 36%, 81%, and 65% respectively). Briganti nomogram (OR = 1.03, p  = 0.04) and 18 F-fluciclovine visual score (≥ 4) (OR = 4.27, p  = 0.02) resulted independent predictors of LNM at multivariable analyses. On region-based semi-quantitative analyses (n = 576), PET/CT performed better using TBR parameters (TBR-A similar to TBR-BM; TBR-A fluciclovine AUC 0.61, p 0.35, vs choline AUC 0.57 p 0.54; TBR-BM fluciclovine AUC 0.61, p 0.36, vs choline AUC 0.58, p 0.52) rather than using absolute LN SUVmax (fluciclovine AUC 0.51, p 0.91, vs choline AUC 0.51, p 0.94). However, in all cases, diagnostic performance was not statistically significant for LNM detection, although slightly in favor of the experimental tracer 18 F-fluciclovine for each parameter. On the contrary, visual interpretation significantly outperformed PET semi-quantitative parameters (choline and fluciclovine: AUC 0.65 and 0.64 respectively; p 0.03) and represents an independent predictive factor of LNM with both tracers, in particular 18 F-fluciclovine (OR = 8.70, p 0.002, vs OR = 3.98, p  = 0.03). Conclusion In high-risk primary PCa, 18 F-fluciclovine demonstrates some advantages compared with 11 Ccholine but sensitivity for metastatic LN detection is still inadequate compared to PLND. Visual (combined morphological and functional), compared to semi-quantitative assessment, is promising but relies mainly on readers’ experience rather than on unquestionable LN avidity. Trial registration EudraCT number: 2014–003,165-15