ABSTRACT. Da ta fr o m Columbi a Gl ac ier a re used to id e ntify processes t ha t contro l ca king from a tem pe ra te tid ewa te r g lac ier a nd to rc-eva lu ate modcls th a t h a \'e bee n proposed to desc ribe ice berg ca king. S in ce 198 1, Co lumbi a Gl ac ier h as b een retrea ting ra pidl y, with a n a lm os t seve n-fo ld increase in calv in g ra te fro m th e mid1970s to 1993 . At th e sa m e time, th e speed of th e g lac ier inc reased a lm ost as mu c h , so th a t th e ac tu a l ra te of retrea t in creased m o re slowl\' . Acco rdin g to th e comm o nl y acce pted m odcl , the ca lving ra te is Lin ea rl y rel a ted to th e wa te r d e pth a t th e te rminu s, \I'i th retrea t of th e glac ier sno u tin to dee per w a ter, leadi ng to I a rge r cah 'in g ra tes a nd acce lera ted r e trea t. Th e Co lumbi a G lac ie r d a ta show th a t th e ca kin g ra te is no t simpl y link ed to observed qu a ntiti es such as wa te r de pth o r s tre tchin g ra te nea r th e terminu s. Durin g th e retreat, th e thi ckn ess a t th e terminu s a ppears to be lin ea rl y correla ted with the wa ter d e pth ; a t th e te rminus, th e thi c kn ess in excess of fl o ta ti o n rem a in ed a t a bo ut 50 m . Thi s suggests th a t re trea t may be initi a ted \I'hen th e te rmin a l thi ckn ess b eco m es too sma ll , vvith th e ra te of re treat contro ll ed b y th e ra te a t w hi c h th e snout is thinnin g a nd by th e basa l slope. Th e impli cation is t h a t th e ra pid re trea t of Co lumbi a Glac ier (a nd o th e r co mpa ra ble ti devva ter glac iers) is no t th e res ult o f a n in crease in ca lving as th e g lac ier re trea ted into d eeper wa te r. r ns tead , th e re trea t was initi a ted a nd m a inta ined by thinning of th e g lac ier. Fo r Co lumbi a G lac ie r , th e continu ed thinning is proba bl y assoc ia ted with th e in crease in glac ier speed a nd ret rea t m ay be ex pected to co ntinue as lo ng as th ese la rge sp eed s a rc main ta in ed. It is no t clea r wh a t mec ha nism m ay be respo nsibl e fo r th e speed-up but the m os t lik ely candid a te is a c ha nge in b asa l co nditi ons o r subg lacia l dra in age. Co nseq ue ntl y, th e beha\' io r o f tid ewa ter glac ie rs m ay be co ntro ll ed b y processes ac tin g a t th e g lacie r bed ra th er th a n b y wha t ha ppe n s a t th e glac ier te rminu s.