Background: Power analysis and sample size planning are key components in designing cluster randomized trials (CRTs), a common study design to test treatment effect by randomizing clusters or groups of individuals. Sample size determination in two-level CRTs requires knowledge of more than one design parameter, such as the effect size and the intraclass correlation (ICC). Their true values, nonetheless, are, by definition, unknown. One classical approach, the face value approach, is to provide the parameter estimates taken at their face values from previous studies (Anderson et al., 2017). However, ignoring the uncertainty inherent in the parameter estimates can lead to optimistic estimation of the required sample size, hence underpowering a study in design (Anderson et al., 2017; McShane & Böckenholt, 2016). Another approach, the safeguard power approach, is to provide conservative estimates of the parameters (Perugini et al., 2014). This approach can at times be overly conservative and suggest a lot more samples than needed (McShane & Böckenholt, 2016; Pek & Park, 2019). For single-level design, an alternative approach to incorporate uncertainty for power analysis is the hybrid classical-Bayesian approach (Spiegelhalter et al., 2004). Studies have shown that this approach provides more realistic power estimates than the classical approaches (McShane & Böckenholt, 2016; Pek & Park, 2019). As demonstrated in McShane & Böckenholt (2016), this approach realistically suggests more than 300 samples to achieve 80% power on average with a modest amount of uncertainty in the effect size, whereas ignoring uncertainty underestimated more than half the required sample size. Despite the significance of the hybrid classical-Bayesian approach in single-level design, to our knowledge, it has not been developed for multilevel designs. Purpose: To that end, the aim of the present study is to extend the hybrid classical-Bayesian approach to two-level CRTs for power analysis and sample size planning. Specifically, we incorporate the uncertainty in the effect size and ICC estimates into power analysis for more realistic sample size determination.