This paper compares the efficacy of covariance-based SEM (CBSEM) and variance-based SEM (PLS) through a large-scale Monte Carlo simulation. The study aims to provide empirical evidence on the relative performance of these two approaches under various conditions, addressing the lack of quantitative guidelines for researchers. Key findings include: 1. **Parameter Consistency and Accuracy**: CBSEM outperforms PLS in terms of parameter consistency, especially when sample sizes are large (over 250 observations). PLS is more accurate in parameter estimation, but only when sample sizes are small (less than 250 observations). 2. **Statistical Power**: PLS generally has higher statistical power than CBSEM, making it more suitable for prediction and theory development. For example, PLS can achieve acceptable statistical power with fewer observations than CBSEM. 3. ** Robustness**: CBSEM is robust to violations of its underlying distributional assumptions, while PLS does not require any distributional assumptions but may suffer from inconsistent parameter estimates if the number of indicators per construct and sample size are not sufficiently large. 4. **Choice Between Methods**: The choice between CBSEM and PLS should consider the research focus. CBSEM is preferred for confirming theoretically assumed relationships, while PLS is better for prediction and theory development. 5. **Design Factors**: The study examines four design factors: number of indicators per construct, sample size, distribution, and indicator loadings. These factors significantly influence the performance of both methods, with sample size being the most influential factor. 6. **Conclusion**: The paper provides researchers with a comprehensive framework to choose between CBSEM and PLS based on the specific research objectives and conditions.This paper compares the efficacy of covariance-based SEM (CBSEM) and variance-based SEM (PLS) through a large-scale Monte Carlo simulation. The study aims to provide empirical evidence on the relative performance of these two approaches under various conditions, addressing the lack of quantitative guidelines for researchers. Key findings include: 1. **Parameter Consistency and Accuracy**: CBSEM outperforms PLS in terms of parameter consistency, especially when sample sizes are large (over 250 observations). PLS is more accurate in parameter estimation, but only when sample sizes are small (less than 250 observations). 2. **Statistical Power**: PLS generally has higher statistical power than CBSEM, making it more suitable for prediction and theory development. For example, PLS can achieve acceptable statistical power with fewer observations than CBSEM. 3. ** Robustness**: CBSEM is robust to violations of its underlying distributional assumptions, while PLS does not require any distributional assumptions but may suffer from inconsistent parameter estimates if the number of indicators per construct and sample size are not sufficiently large. 4. **Choice Between Methods**: The choice between CBSEM and PLS should consider the research focus. CBSEM is preferred for confirming theoretically assumed relationships, while PLS is better for prediction and theory development. 5. **Design Factors**: The study examines four design factors: number of indicators per construct, sample size, distribution, and indicator loadings. These factors significantly influence the performance of both methods, with sample size being the most influential factor. 6. **Conclusion**: The paper provides researchers with a comprehensive framework to choose between CBSEM and PLS based on the specific research objectives and conditions.