Although tumors frequently show elevated protease activities, the concept of anti-proteolytic cancer therapy has lost momentum after failure of clinical trials with broad-spectrum matrix metalloproteinase inhibitors. Thus we need to adapt our design strategies for protease inhibitors. Here, we employed a series of seven structurally fine-modulated and pharmacokinetically closely related synthetic 4-amidinobenzylamine-based inhibitors with distinct selectivity for prototypical serine proteases in a murine T cell lymphoma liver metastasis model. This in vivo screening revealed efficacy of urokinase inhibitors but no correlation between urokinase selectivity or affinity and anti-metastatic effect. In contrast, factor Xa-selective inhibitors were more potent, demonstrating factor Xa or a factor Xa-like serine protease likely to be more determinant in this model. Factor Xa selectivity, but not affinity, significantly improved anti-metastatic efficacy. For example, factor Xa inhibitors CJ-504 and CJ-510 exert similar affinity for factor Xa (K(i)=14 nM versus 8.8 nM) but CJ-504 was 70-fold more selective for factor Xa. This correlated with higher anti-metastatic efficacy (58.8% with CJ-504; 28.2% with CJ-510). Our results show that among the protease inhibitors employed that have affinities in the nanomolar range, the strategy of selectivity-optimization is superior to further improvement of affinity to significantly enhance anti-metastatic efficacy. This appreciation may be important for the future rational design of new anti-proteolytic agents for cancer therapy.