Market prices can be off due to microstructure noise, making it hard to estimate volatility accurately. Summing small returns helps, but noise adds up too. Realized volatility can't always show true volatility, but a modified version can reveal noise characteristics. Simple calculations on noisy data can give reliable noise estimates. Sampling high-frequency data optimally can balance bias and variance in volatility estimation.