Treat­ment de­sign for mus­cu­loskele­tal dis­or­ders us­ing in sil­ico pa­tient-spe­cific dy­namic sim­u­la­tions is be­com­ing a clin­i­cal pos­si­bil­ity. How­ever, these sim­u­la­tions are sen­si­tive to model pa­ra­me­ter val­ues that are dif­fi­cult to mea­sure ex­per­i­men­tally, and the influence of un­cer­tain­ties in these pa­ra­me­ter val­ues on the ac­cu­racy of es­ti­mated knee contact forces re­mains un­known. This study eval­u­ates which mus­cu­loskele­tal model para­me­ters have the great­est in­flu­ence on es­ti­mat­ing ac­cu­rate knee con­tact forces during walk­ing. We per­formed the eval­u­a­tion us­ing a two-level op­ti­miza­tion al­go­rithm where mus­cu­loskele­tal model pa­ra­me­ter val­ues were ad­justed in the outer level and mus­cle ac­ti­va­tions were es­ti­mated in the in­ner level. We tested the al­go­rithm with differ­ent sets of de­sign vari­ables (com­bi­na­tions of op­ti­mal mus­cle fiber lengths, ten­don slack lengths, and mus­cle mo­ment arm off­sets) re­sult­ing in nine dif­fer­ent op­ti­miza­tion prob­lems. The most ac­cu­rate lat­eral knee con­tact force pre­dic­tions were ob­tained when ten­don slack lengths and mo­ment arm off­sets were ad­justed si­mul­ta­ne­ously, and the most ac­cu­rate me­dial knee con­tact force es­ti­ma­tions were ob­tained when all three types of pa­ra­me­ters were ad­justed to­gether. In­clu­sion of mo­ment arm off­sets as de­sign vari­ables was more im­por­tant than in­clud­ing ei­ther ten­don slack lengths or op­ti­mal mus­cle fiber lengths alone to ob­tain ac­cu­rate me­dial and lat­eral knee con­tact force predic­tions. These re­sults pro­vide guid­ance on which mus­cu­loskele­tal model pa­ra­me­ter val­ues should be cal­i­brated when seek­ing to pre­dict in vivo knee con­tact forces ac­cu­rately.