The yeast split-ubiquitin system has previously been shown to be suitable to detect protein interactions of membrane proteins and of transcription factors in vivo. Therefore, this technology complements the classical split-transcription factor based yeast two-hybrid system (Y2H). Success or failure of the Y2H depends primarily on the ability to avoid false-negative and false-positive hits that become a limiting factor for the value of the system, especially in large scale proteomic analyses. We provide here a systematic assessment of parameters to help improving the quality of split-ubiquitin cDNA-library screenings. We experimentally defined the optimal 5-fluoroorotic acid (5-FOA) concentration as a key parameter to increase the reproducibility of interactions and, at the same time, to keep non-specific background growth low. Furthermore, we show that the efficacy of the 5-FOA selection is modulated by the plating density of the yeast clones. Moreover, a reporter-specific class of false-positive hits was identified, and a simple phenotypic assay for efficient de-selection was developed. We demonstrate the application of this improved system to identify novel interacting proteins of the human Frizzled 1 receptor. We identified several novel interactors with components of the Wnt-Frizzled signalling pathways and discuss their potential roles as direct mediators of Frizzled receptor signalling. The present work is the first example of a split-ubiquitin interaction screen using an in-situ expressed receptor of the serpentine class, emphasizing the suitability of the described improvements in the screening protocol.