Contralateral audiometric thresholds and speech intelligibility vary greatly across cochlear implant (CI) users with acoustic hearing in the non-implanted ear. This study systematically compares spatial speech-in-noise performance in these (bimodal or single-sided deaf, SSD) CI users using three different paradigms, in order to reveal the influence of spatial scenario and contralateral audibility on speech intelligibility. The three paradigms include (1) measurements of 16 actual CI users with a mobile measurement platform, (2) vocoder simulations of electric and acoustic listening with normal hearing (NH) listeners, and (3) an “effective” model of electric, acoustic and bimodal electroacoustic speech-in-noise performance. All three paradigms controlled for secondary factors, e.g., head movements, room acoustics, and hearing aid signal processing (using the Master Hearing Aid). Eleven NH participants listening to unprocessed speech and noise were used as a reference group. The results show a clear dependence of speech reception thresholds (SRTs) on spatial scenario, and a dependence of both SRTs and spatial release from masking (SRM) on contralateral audibility in all three paradigms. Task-specific “better-ear listening” dominates the results in all three paradigms. This is true even for the effective model, where the possibility of processing the signal’s fine time structure across ears was enabled, but was not effective. Instead, the model utilized the best signal-to-noise ratio of either acoustic or electric listening, which could change, depending on contralateral audibility and spatial scenario. Both, model and simulations were capable of predicting patient data with high accuracy (rank correlation of 0.99 for SSD, and 0.88 for bimodal CI users). Thus, model and simulations create “virtual” SSD and bimodal CI listeners that can be used to evaluate signal enhancement algorithms in a systematic way, before testing them on actual CI users.