diff --git a/Gemfile.lock b/Gemfile.lock
index eba4c11d..e016d71d 100644
--- a/Gemfile.lock
+++ b/Gemfile.lock
@@ -152,4 +152,4 @@ DEPENDENCIES
   test-unit
 
 BUNDLED WITH
-   2.1.2
+   2.1.4
diff --git a/lib/simplecov/source_file.rb b/lib/simplecov/source_file.rb
index 4815c494..4cb9f632 100644
--- a/lib/simplecov/source_file.rb
+++ b/lib/simplecov/source_file.rb
@@ -199,7 +199,7 @@ def read_lines(file, lines, current_line)
       return lines unless current_line
 
       set_encoding_based_on_magic_comment(file, current_line)
-      lines.concat([current_line], file.readlines)
+      lines.concat([current_line], ensure_remove_undefs(file.readlines))
     end
 
     RUBY_FILE_ENCODING_MAGIC_COMMENT_REGEX = /\A#\s*(?:-\*-)?\s*(?:en)?coding:\s*(\S+)\s*(?:-\*-)?\s*\z/.freeze
@@ -211,6 +211,15 @@ def set_encoding_based_on_magic_comment(file, line)
       end
     end
 
+    def ensure_remove_undefs(file_lines)
+      # invalid/undef replace are technically not really necessary but nice to
+      # have and work around a JRuby incompatibility. Also moved here from
+      # simplecov-html to have encoding shenaningans in one place. See #866
+      # also setting these option on `file.set_encoding` doesn't seem to work
+      # properly so it has to be done here.
+      file_lines.each { |line| line.encode!("UTF-8", invalid: :replace, undef: :replace) }
+    end
+
     def build_lines
       coverage_exceeding_source_warn if coverage_data["lines"].size > src.size
       lines = src.map.with_index(1) do |src, i|