This　paper　presents　a　conjugate　gradient　algorithm　of　convergence　speed　and　accuracy　in　pipelined　ADCs.　The　pipeline　ADC　error　acts　as　the　objective　function　of　the　method.　The　objective　function　gradient　gradually　produce　conjugate　direction　which　acts　as　the　search　direction　to　obtain　the　minimum　point,　where,　the　search　direction　is　just　a　combination　of　the　negative　gradient　direction　with　the　last　iteration　of　the　search　direction.　It　takes　only　a　first　derivative　information　to　overcome　the　slow　convergence　shortcoming　of　the　steepest　descent　method.　Simultaneously,　it　can　also　avoid　the　disadvantage　of　storing　data　and　computing　Hesse　matrix　in　the　Newton　method.　Simulation　results　shows　that　the　error　converge　to　minimum　point　in　a　short　time.　The　proposed　calibration　has　a　small　amount　of　memory　required　and　high　stability.