Earnings management is the choice by a manager of accounting policy to achieve some specific objectives; such choices divide into two categories. One is choice of accounting policies, such as straight-line versus declining-balance amortization, or policies for revenue recognition. The other is discretionary accrual, such as provision for credit losses, warranty costs, inventory values, and timing and amounts of non-recurring and extraordinary items like write-offs and provision for recognization (Scott, 2003). An establishment wanting to manage earnings has many tricks to choose, such as adjusting operational discretionary accruals or arranging dealings with its conglomerates. Many scholars study this field, using discretionary and non-discretionary accruals to detect earnings management. Yet few studies confer on related party transactions as a way of earnings management. Actually, there are many instances of controlling stockholders pillaging their own firms through related party transactions. To transfer assets and earnings out of firm for the benefit of those who control them, such as transactions between Enron, Worldcom and its special-purpose entities, firms conduct related party transactions for two reasons. One is to minimize transaction costs, as normal related party transactions. The other is to use related party transactions to manage earnings, as abnormal related party transactions and related party sales as a key related party transaction item for earnings management. Firms could dissimulate opportunistic behavior by manipulating related party sales, since such sales figures are disclosed only in footnotes, not in the income statement (Jian and Wong, 2004). Hence it is not easy to distinguish between normal and abnormal related party sales. The major purpose of the study is that we are a pioneer in detecting abnormal related party sales (RPS) model, considering recent financial frauds in Taiwan. In the case of manipulating company financial reports, it was revealed that an entity could operate well and attain fiscal solvency. It did much trade with its conglomerates and cashed accounts receivable at the end of the fiscal year, such that the firm could raise funds from investors successfully. Related party transactions are in want of operating for a business once in a while. If an entity makes bargains with its conglomerate persistently, motives of the transactions are dubious, and it is so perplexing to distinguish between normal or abnormal related party transactions. Although specific disclosure requirements are set forth in Statement on Auditing Standards (SAS) Number 6, which provides guidance to auditors in identifying related party transactions, the abnormal related party transactions on financial statement in fact could not be presented in the integrity. So it is a common manner of managing earnings by making bargains with conglomerates in Taiwan. For this reason, we develop the RPS Model to detect abnormal related-party transactions to atone for sad fact that financial reporting is imperfect. And we evaluate performance of the RPS Model by contrasting specification and power, using test statistics. The RPS Model reveals specific items well when applied to a random sample of firm-years, and it also proves that the fine power of RPS Model, measured by cross-sectional method. Managerial discretion hypotheses are presented from existing literature, and an accrual-based model is used for detecting earnings management, such as Jones Model (Jones, 1991) and Modified Jones Model (Dechow et al., 1995).They estimate the discretionary components of reported income. Those models appear highly specific and good for detecting manipulated earnings. The accuracy of the Modified Jones Model (Dechow et al., 1995) rises to a hundred percent when earnings manipulation amount come to 10 % of prior assets. But an industry may do extensive trade with conglomerates and cash accounts receivable at the end of the fiscal year for the sake of diminishing the discretionary accruals. Actually such a business manipulates related party transactions for its own interests; it is a pity that those accrual-based models could not detect the management level effectively. Low accuracy of accrual-based models means the model is invalid. It is our belief that there is no use detecting earnings management which administrative department diminishing discretionary accruals items with accrual-based models. It is the restrictions on accrual-based models whereby discretionary accruals are diminished. In this research the RPS Model uses selling expenses and administrative expenses to project the abnormal sales amount from abnormal related party transactions. Management manipulates earnings without respect to accrual or non-accruals; the RPS Model could detect earnings management amount more effectively than accrual-based models. To an extent, our assumptions are that selling expenses and administrative expenses project the normal sales amount and multiply by the ratio of related party sales is normal related party sales amount. Subtract this from actual relatedparty sales amount and we have abnormal amount of related-party sales. Thus, abnormal accruals and amount of nonaccrual-basis related-party sales are all included in our estimated abnormal amount of related-party sales. Results show RPS Model as an accurate specification via random sample. Besides, it also proves the accuracy of the RPS Model, measured by cross-sectional way, almost reaches one hundred percent, while the amount of related party transactions rises to nearly four percent of prior total assets. So the RPS Model could detect abnormal related-party sales effectively and usefully.