Background: In hospital- and community-acquired central nervous system infections, resistant Gram-positive bacteria are an increasing therapeutic challenge. The present approach does not attempt to identify rapidly bactericidal therapies for susceptible pathogens but aims to improve methods to find antibiotic regimens for multi-resistant pathogens that are effective in vivo in spite of reduced in vitro susceptibility in culture media. Methods: Antibiotic susceptibility was tested in cerebrospinal fluid (CSF) and Mueller–Hinton broth (Enterococcus faecalis, methicillin-resistant Staphylococcus aureus, Staphylococcus epidermidis) or brain–heart infusion (Streptococcus pneumoniae). Results: Minimal inhibitory concentrations (MICs) and minimal bactericidal concentrations (MBCs) were either lower in CSF than in broth or equal in CSF and broth. The difference between MICs in CSF and broth was prominent with gentamicin, levofloxacin, linezolid (staphylococci), and vancomycin (staphylococci and pneumococcus), whereas it was absent with ampicillin (E. faecalis), penicillin G (S. pneumoniae), linezolid (enterococcus and pneumococcus), and vancomycin (enterococcus). In no case was the MIC or MBC higher in CSF than in broth. Conclusions: Several antibiotics possess an antibacterial effect in CSF at lower concentrations than the MICs determined in broth, i.e., MICs in broth underestimate in situ susceptibility in CSF.