Social scientists constantly face the challenge of recovering information from messy and/or limited data. This article advocates a backward-elimination approach to cross-level inference that takes full advantage of known auxiliary data to recover unobserved and yet crucial information. Through examining the controversial case of the Legislative Yuan’s Consent Vote to the Nominees of the President and Vice-President of the Examination Yuan on June 21,2002, we show how to solve the puzzle of “who abstained in the vice-president nominee’s consent vote” by inferring the voting patterns of each party’s legislators from aggregate vote counts and news reports. Our solution is quite robust to various assumed scenarios despite the contrary claims made by politicians and pundits. We thus conclude that political scientists can expand their academic landscape by making creative use of this rich arsenal of research methodology.