This paper examines the relationship between gross returns and the bid-ask spread, extending Amihud and Mendelson's (1986) model to incorporate both the clientele effect (concave relation) and Jacoby, Fowler, and Gottesman's (2000) level effect (convex relation). The extended model predicts that the relationship is concave for highly liquid assets with low relative spreads and convex for thinly traded assets with higher relative spreads. Using NASDAQ data, the authors test this hypothesis and find evidence supporting the extended model. The results show that the clientele effect dominates the level effect for highly liquid assets, while the level effect dominates for thinly traded assets. The paper concludes that highly liquid NYSE stocks tend to lie in the concave range, where the clientele effect is dominant, while a significant number of less liquid NASDAQ stocks lie in the convex range, where the level effect is dominant. Additionally, controlling for convexity, a linear relation between returns and beta is identified, supporting the argument that adjusting the CAPM for liquidity effects can eliminate nonlinearity.This paper examines the relationship between gross returns and the bid-ask spread, extending Amihud and Mendelson's (1986) model to incorporate both the clientele effect (concave relation) and Jacoby, Fowler, and Gottesman's (2000) level effect (convex relation). The extended model predicts that the relationship is concave for highly liquid assets with low relative spreads and convex for thinly traded assets with higher relative spreads. Using NASDAQ data, the authors test this hypothesis and find evidence supporting the extended model. The results show that the clientele effect dominates the level effect for highly liquid assets, while the level effect dominates for thinly traded assets. The paper concludes that highly liquid NYSE stocks tend to lie in the concave range, where the clientele effect is dominant, while a significant number of less liquid NASDAQ stocks lie in the convex range, where the level effect is dominant. Additionally, controlling for convexity, a linear relation between returns and beta is identified, supporting the argument that adjusting the CAPM for liquidity effects can eliminate nonlinearity.