Using low-frequency data, the high-low spread estimator performs significantly better than the Roll covariance estimator. This paper expands upon this study to explore several unanswered issues. Why does the high-low spread estimator outperform the Roll estimator when low-frequency data is employed, and how is the performance of the high-low spread estimator computed from high-frequency data? Is it still better than the Roll estimator? To shed light on these questions, a trading day is divided into several intraday periods to create a greater number of high-low ratios in order to develop a spread estimator, which is referred to as a fractional high-low spread estimator. Analysis indicated that the fractional high-low spread estimators have better performance when the true spread is even wider, when the frequency of transaction prices becomes even higher, when price volatility is relatively lower, and when asset prices tend to reverse. Moreover, the fractional high-low spread estimator broadly outperforms the Roll estimator when estimators are computed from a dataset with the length of an intraday period longer than 30 minutes. In contrast, the Roll estimator generally surpasses the fractional high-low spread estimator, when the length of the intraday period is shorter than 20 minutes.