The sum of squared returns, or realized volatility, of the recently available high frequency financial data should be a good estimator for integrated volatility. However, the empirical studies suggest that the market microstructure noise which contaminates the efficient prices would make realized volatility inconsistent. We review the recent literature on the estimators for integrated volatility and the market microstructure noise. We focus on the statistical properties and the empirical findings of the subsample-based estimators, e.g., Two Scales Realized Volatility (TSRV), and the kernel-based estimators. Our empirical analysis on the transactions of two actively-traded Taiwan Stock Exchange stocks does not reject the assumption that the market microstructure noise is serially independent and independent of the efficient price. Our results support that TSRV is practically applicable for computing realized volatility of these two stocks. We also find that the jumps in the efficient prices might not be negligible by bipower variation. We propose a method to estimate the autocorrelation of the noise by the empirical autocorrelation of the log-returns.