The  education  of  the  poor  and  disadvantaged  population  has  been  a  long-standing  challenge  to  the  education  system  in  both  developed  and  developing  countries.  Although computer assisted  learning  (CAL)  has  been  considered  one  alternative  to  improve  learning outcomes  in  a  cost effective  way,  the  empirical  evidence  of  its  impacts on  improving  learning outcomes  is  mixed.  This  paper  intends  to  explore  the  nature  of  the  effects  of  CAL  on  student academic  and  non­‐academic  outcomes  for  underserved  populations  in  a  developing  country. To meet  this  goal,   a  randomised  field  experiment  of  a  CAL  programme  involving  over 4000  third grade  students,  mostly  from  poor  migrant  families,  in  43  migrant  schools  in Beijing was conducted. The  main  intervention  is  a  math  CAL  programme  that  is  held  out  of  regular  school hours. The  programme  is  tailored  to  the  regular  school  math  curriculum  and  is  remedial  in nature. The results  show  that,  the  CAL  programme  improved  the  student  standardised  math scores by 0.14 standard  deviations  and  most  of  the  programme  effect  took  place  within  two  months after the  start  of  the  programme.  Low­‐performing  students  and  those  with  less educated parents benefited  more  from  the  programme.  Moreover,  CAL  also  significantly increased the  levels  of  self­‐efficacy  of  the  students  and  their  interest  in  learning.  At  most a moderate  programme  spillover  in  Chinese  test  scores was observed.  Our  findings  are  robust to  the Hawthorne effect  and  CAL  programme  spillovers  that  might  potentially  bias  the  estimates of  the programme effects.