diff --git a/book/knn.tex b/book/knn.tex
index 3833e6d..1a0ff81 100644
--- a/book/knn.tex
+++ b/book/knn.tex
@@ -208,7 +208,7 @@ \section{From Data to Feature Vectors}
 
 \Figure{knn_classifyitbad}{A figure showing an easy NN
   classification problem where the test point is a ? and should be
-  positive, but its NN is actually a negative point that's noisy.}
+  negative, but its NN is actually a positive point that's noisy.}
 
 
 A solution to this problem is to consider more than just the single