This study evaluates comparatively the effect of various outlier treatment methods on explanatory power and earnings response coefficient in estimating the relation between unexpected earnings and abnormal returns under a linear regression model. Existing studies treat outliers in unexpected earnings generally by truncating or winsorizing within ±1% of its distribution. Such methods, however, are weak in enhancing model explanatory power and earnings response coefficient. This study shows that model explanatory power can be substantially improved by raising the outlier cutoff point to ±5% or by ranking unexpected earnings entirely. ±5% outlier treatment also has enhanced earnings response coefficient markedly. Including earnings level in addition to unexpected earnings in the model has further improved model explanatory power and earnings response coefficient. Without outlier treatment, the model explanatory power(Adj. R²) was merely 0.041 on average. Winsorizing or truncating at ±1% increased Adj. R² to 0.088 and 0.089 respectively. Ranking, however, increased it substantially to 0.145. Further, ranking after adding earnings level in the model increased it markedly to 0.241. Winsorizing or truncating at ±5% was similarly effective in increasing Adj. R² as ranking. Without outlier treatment, earnings response coefficient was merely 0.091. Winsorizing or truncating at ±1% increased earnings response coefficient to 0.273 and 0.416 respectively. Winsorizing or truncating at ±5%, however, increased it markedly to 0.949 and 1.217 each. Lev(1989) expresses concern about a weak link between earnings and stock returns, Adj. R² being merely 0.05, documented in existing earnings information content studies. The results of this study indicate that strengthening outlier treatment and including earnings level in addition to unexpected earnings in the model can substantially mitigate such worry. The results also show that strengthening outlier treatment can enhance earnings response coefficient up to its minimum theoretical value, 1.