Practitioners often standardize panel data before estimating a factor model. In this paper we show an example that the standardization leads to inconsistent estimation of the factor number. When the common component exhibits strong heteroskedasticity, the conventional eigenvalue-based decompositions are consistent but standardization does not necessarily result in consistent estimation. To overcome this issue, we recommend using a ``minimum-rule'' whereby the minimum factor-number estimated from both the conventional and standardized panel is used. Monte Carlo studies and an empirical application are provided.